麻豆传媒’s explainer on what MCP is covers the standard and what a connection looks like. This picks up from there and goes through what has actually been built.

An “official” MCP server is built and maintained by the platform vendor itself, the way Salesforce, HubSpot or 麻豆传媒 build and support theirs. A “community” server is built by an independent developer, and it may or may not be maintained over time. Think of it as the difference between a manufacturer’s own accessory and an aftermarket part: both can do the job, and only one comes with somebody obliged to answer the phone.

Both are useful. You just judge them differently, and the section at the end covers how.

Where a vendor publishes its own documentation, the links below go there. Where a server exists only as a community project, the link goes to a registry listing, which is a place to start looking rather than a recommendation.

MCP servers for construction and design platforms

Autodesk publishes the most developed set of official servers in this category, though the coverage sits on the design side rather than the field side. Its is fully released and gives an assistant read-only access to documentation across more than 110 products, and a Fusion Data server covers Fusion’s backend project and folder data. Two more are earlier in the cycle: the is in tech preview and a Fusion Automation server is in private beta, so treat both as worth watching rather than worth building on this quarter. For teams working with Autodesk 麻豆传媒 Services, Autodesk also publishes against its APIs, which is the realistic path if what you need is Autodesk Build or Construction Cloud data rather than model data.

Procore does not currently publish its own MCP server for project data, though it is building MCP support directly into the platform. Its now lets apps declare MCP servers and AI agents as components of a Procore app, with the stated aim of making every Marketplace app agent-ready. In the meantime several community servers exist against the Procore API, and Zapier offers a Procore connector. Anyone evaluating a community option should look closely at how it authenticates and what access it asks for, because Procore holds a great deal of commercially sensitive project information.

CMiC and Unanet have no official server and no established community server that we were able to verify, which leaves anyone on those platforms building against the API directly if they want a connection at all.

In practice that means a question like “what changed in the mechanical model since last week” is reachable today through Autodesk’s design-side servers, whereas a question like “which RFIs on this job are still open and who owns them” is not, because it lives in the project record rather than the model. That is the half nobody has shipped, and it is the half most field teams would actually use.

MCP servers for CRM and pursuit data

This is the most mature category in the list, and it matters more for construction than it first appears, because pursuit data is what turns workforce planning from reactive into forward-looking. Knowing a job sits at seventy percent probability three months before award is the difference between staffing it deliberately and scrambling for a superintendent the week it signs.

An assistant that can reach the pipeline answers questions that currently require somebody to pull a report and cross-reference it by hand, and those questions get asked far less than they should precisely because of how long they take to answer.

Salesforce supports MCP through Agentforce, with covering how resources, tools and prompts are exposed to an assistant. It works through your existing logins, so an assistant can only see what the person using it could already open.

HubSpot ships , one connecting any MCP-compatible assistant to your CRM for querying and writing records, and a second serving developers building on the HubSpot platform. Both connect through your existing HubSpot login, so access follows whatever that account can already reach.

Microsoft Dynamics 365 publishes official servers for and for Customer Service, the latter generally available since July 2026, with a Commerce server in preview. Supported clients include Copilot Studio alongside external ones such as Claude Code, ChatGPT and VS Code. For contractors already standardised on Microsoft, this arrives inside a governance model the IT team has almost certainly configured already.

Community and emerging servers

• Zoho Bigin. Simplified pipeline CRM with a community server via Smithery, relevant to smaller contractors already running the Zoho suite.
• Pipedrive. Deal-tracking CRM with a community server, common among regional contractors and specialty firms that never adopted an enterprise CRM.

MCP servers for HR and people data

This category is close to empty.

We could not verify an official MCP server for Workday, BambooHR, UKG, Oracle, ADP or Namely. Workday’s developer portal makes no mention of the protocol at all, and a URL that looks like a BambooHR MCP endpoint resolves to a login page rather than to documentation.

For workforce planning this is the most consequential absence on the list, because HR systems hold the roster, the certifications, the cost codes and the start dates that make a staffing answer trustworthy. Without a connection, an assistant works from whatever somebody last exported and uploaded, and the export is where this goes wrong. A roster pulled in March will answer a question about June with complete confidence and no indication that three people have left and two certifications have lapsed, which is a worse outcome than getting no answer at all.

Until those vendors ship, the practical route is a workforce platform that already syncs from your HRIS and exposes its own connection layer, so the assistant reads current state rather than a snapshot. That is the approach 麻豆传媒 takes with its MCP connection to live workforce data.

MCP servers for spreadsheets, data and warehouses

Airtable publishes available on all plans, with access matching whatever the person connecting it can already see, which makes it useful for the trackers and light databases that accumulate around a construction operation.

Google BigQuery ships that interprets your schema and translates natural-language questions into SQL without moving data out of the warehouse, which is relevant to any contractor whose reporting layer already sits there.

Snowflake publishes MCP capability through its Cortex tooling, though the documentation has moved recently enough that it is worth navigating from the Snowflake docs homepage rather than trusting a link from an article, this one included. Warehouse connections tend to be the ones IT reviews most carefully, so budget time for that conversation.

MCP servers for construction document storage

Construction runs on documents more than it runs on anything else, and the file-storage category is comfortably the best covered on this list.

Dropbox offers connecting an assistant to your stored files, which is the simplest starting point on this whole list. Box publishes an official server aimed at enterprise content management, which matters for firms with retention and compliance requirements on project records. Microsoft publishes MCP references covering OneDrive and the wider Microsoft 365 surface, which is the relevant path for the large share of contractors running on SharePoint.

MCP servers for coordination and communication tools

Slack publishes covering channel and message access, useful mainly for reconstructing what was decided in a thread nobody minuted. Asana offers for task and project data, and monday.com publishes . Atlassian ships covering Jira and Confluence, which appears more in construction technology teams than in the field. Notion publishes for the teams keeping their standards and process documentation there.

None of these are construction tools, and every one of them is a place construction coordination actually happens.

Zapier publishes exposing thousands of connected applications through a single server. For a system with no MCP of its own but an existing Zapier integration, this is the pragmatic bridge, at the cost of an extra hop and a dependency on how well that Zapier integration was built.

Which MCP servers contractors should connect first

If you are starting from nothing, the order below gets you working connections without waiting on anybody’s roadmap.

Begin with file storage, whichever of SharePoint, Dropbox or Box holds your project documents. It is the highest-volume, lowest-risk connection available and it answers the largest share of everyday questions, which are mostly people trying to find what a document said.

Move next to the CRM, because pursuit data is the input that makes forward-looking staffing questions answerable at all, and because the servers in that category are the most mature on this list.

Add your data layer third, meaning Airtable, BigQuery or whatever holds the trackers and reports that accumulate around operations. This is where the questions that currently require somebody to build a pivot table start getting answered conversationally.

Leave communication tools until you have a specific reason, because connecting Slack sounds appealing and mostly produces summaries nobody asked for unless you already know the question you want it to answer.

Treat project management and HRIS as pending rather than absent, and ask the vendors directly rather than assuming. That conversation is worth having now even though today’s answer is mostly no.

How to evaluate an MCP server before connecting it

Four questions, ordered by how much trouble each one saves you later.

Who built it, and are they still maintaining it? Check the last commit or the last documentation update, because a community server nobody has touched in a year is a liability rather than a shortcut and the staleness is rarely advertised on the listing.

How much access does it ask for? A well-built server gives the assistant exactly the access the person using it already has, and nothing beyond that. If a server asks for more than that, sort it out before you install rather than after.

Is it read-only or can it write? Read access to a project record carries a very different risk profile from write access, and a surprising share of genuinely useful workflows need only the former, so start read-only wherever the option exists and add write access when something specific requires it.

Who owns it once it is connected? This one gets skipped. A connection nobody owns is a connection nobody reviews, and the review is what catches the access that made sense in March and does not now.

For discovery, the is the canonical index, and community marketplaces including Smithery, mcp.so, Glama and PulseMCP catalogue far more servers with varying quality signals. Listings on those marketplaces go stale, and a dead listing is common enough that it is worth verifying a server exists before planning around it.

What MCP adoption looks like for construction today

The connections available to a contractor today cluster around the business systems: your CRM, your file storage, your warehouse, the tools your office already runs on. Construction-specific servers are arriving as the industry adopts AI, and several are being built right now by vendors and by independent developers working against platform APIs.

So the practical move is to start where connections already exist and build the habit while the rest arrives. Working out who approves a connection, how much access it gets and who reviews it later takes an organisation longer than the technical setup does, and it is the same work whichever system you eventually point it at.

The post Top MCP servers for construction teams and what to connect first appeared first on 麻豆传媒.

A Gem is a saved version of Gemini configured for one job, and the useful ones are narrower than most people expect when they first build one. I want to work through the whole thing here: what a Gem actually holds, how to tell which of your tasks deserve one, how to write instructions that change the output instead of merely describing what you want, what happens the moment you share one, and the point at which you should be reaching for NotebookLM instead.

What a Gem is

A Gem is a saved configuration of Gemini made up of a name, a set of instructions, and optional knowledge files. Google’s own overview describes Gems as a way to from Gemini, personalized to any topic, which undersells how specific the genuinely useful ones are.

Open a Gem and it already holds the instructions you wrote and the files you attached, so the conversation starts with your terminology, your format and your constraints in place rather than needing to be told again. There is no practical limit on how many you build, and the sections below argue for several narrow ones over a single Gem attempting to cover everything you do.

Which tasks deserve a Gem

The test is repetition against a stable standard, so a task you do most weeks measured against a format or rule set that does not change much is a Gem, while everything else is better handled in a normal chat where you can see exactly what you are feeding it.

The instinct most people bring is to build one assistant that knows about the company. That assistant then has to be told every single time which of the many things it knows is relevant right now, which puts you back to explaining context to something you configured specifically so you would not have to.

Narrow works better because instructions serving one task can be specific in ways general instructions cannot. A Gem that only reviews submittals can be told to always cite the spec section, since every question it will ever get is a submittal question, whereas a Gem that also drafts client emails and summarizes meetings cannot be told that without the instruction being wrong two-thirds of the time.

Building your first three Gems

1. The submittal reviewer

Give it the governing spec sections and a review checklist, then constrain what it is allowed to conclude. The constraint is the important part, because an assistant left to itself will produce a verdict, and a verdict is exactly what a coordinator is not authorized to give.

Example instructions: “You review mechanical submittals against the project specification for a specialty contractor. Always cite the spec section you are relying on, quoting the relevant line. List what the submittal covers completely and what it leaves unanswered, in that order. Never approve or reject a submittal and never say whether it will be accepted, because that is the design team’s call. If the spec is ambiguous on a point, say that it is ambiguous rather than choosing the likelier reading.”

2. The look-ahead drafter

Attach a look-ahead you have already sent, because demonstrating the format works better than describing it, and tell the Gem to match it exactly. What belongs in a look-ahead is a solved question at most companies, so the Gem’s job is to follow your answer rather than invent its own.

Example instructions: “You draft our weekly look-ahead from crew assignments and schedule changes. Match the structure of the attached example exactly, including section order and how hours are grouped. Flag any crew assigned to overlapping jobs before anything else in the output. Never invent an hours figure. If a date is missing, list it as outstanding rather than estimating it.”

3. The RFI writer

Attach two or three RFIs you consider well written, because examples carry more than description here: tone is most of what makes an RFI land well, and tone is close to impossible to specify in the abstract. If the people writing them are unclear on the difference between RFIs and submittals, the Gem will inherit that confusion.

Example instructions: “You draft RFIs in our company format, following the attached examples for structure and tone. State the question plainly in the first line. Reference the drawing number and spec section the question arises from. Keep the tone neutral and factual, with no argument about responsibility or cost. If the question could be read two ways, split it into two RFIs.”

4. Write instructions with all four parts

Across all three, the instructions that work contain the same four elements, and the fourth is the one people leave out.

Start with the role, stated narrowly enough to exclude work you do not want it doing. Add the standard it works against, which in construction means naming a specific document rather than appealing to a general principle. Specify the output shape, because “summarize this” and “list what is complete and what is missing, in that order” produce very different things. Finish with the refusals, the two or three moves it must never make, which is the part that gets skipped and the part that matters most on a jobsite. An assistant with no stated refusals will fill gaps confidently, since filling gaps is what it was built to do. Telling it never to estimate an hours figure it was not given, never to approve or reject, and never to resolve a spec ambiguity by picking the likelier reading is what turns a tool producing plausible answers into one producing checkable ones.

Sharing, and the fact that a Gem is a Drive object

When you share a Gem it goes into Google Drive, and is explicit that it runs on the same technology powering Drive, that the experience matches sharing a Doc, and that shared Gems are stored in Drive. Recipients can then edit it, use it as it stands, or take a copy of their own to modify.

For a contractor already standardized on Workspace this is genuinely convenient, since there is no new permissions model to learn and a Gem obeys the file-sharing posture the company already set. It cuts the other way too, and the same announcement says so: if your organization permits sharing documents outside the company, Gems can travel outside the company in exactly the same way. Administrators who want to bound that can control it centrally in the admin console, under Generative AI, then Gemini app, then Gems.

Attaching knowledge from Drive rather than uploading a file has an advantage worth the extra click, since a Drive file the Gem points at is the file your team already maintains, so updating the source updates what the Gem works from while an uploaded copy stays frozen at the moment you attached it.

Two practical notes are worth having before you share anything with colleagues. A Gem is only shareable when its custom knowledge came from a device upload or from Drive, so attaching anything else greys out the share button with no explanation on screen, which is worth knowing before you spend an afternoon wondering what you did wrong. You should also decide deliberately whether your team edits one shared Gem or takes copies, because a single shared Gem stays consistent and accumulates everyone’s improvements, whereas copies drift into four versions that started identical and now behave differently in ways nobody has written down.

When to use NotebookLM instead

Gems draw on Gemini’s full capability plus whatever you attached, which is what makes them good at producing work and wrong for questions you need to defend. For those, Google’s other tool is the right one: NotebookLM answers only from documents you give it and shows the passage each answer came from. 麻豆传媒’s guide on answers grounded in your own documents covers how that works and where it still needs checking.

The split is simple enough to hold in your head: reach for a Gem when you are producing something and already know what good looks like, and reach for NotebookLM when you are asking what a document actually says and the answer has to be traceable back to the page.

Where this falls short

Five Gems is five things to maintain, five instruction sets somebody may have edited, and five sets of knowledge drifting from their sources at their own pace. Whoever builds them should own them, since a shared Gem with no owner becomes something several people have quietly modified in different directions.

A knowledge file inside a Gem is also a copy taken on the day you attached it, so the crew list you upload on Monday is wrong by Thursday if somebody moved, and nothing in the Gem will tell you that. That is the line between a setup that drafts your paperwork well and a system holding the actual state of your workforce, which is what purpose-built AI workforce planning is for. The sensible division is to use Gems for the writing that surrounds crew work while keeping the roster itself somewhere that stays current on its own.

Opinion: build one, not five

The temptation once this clicks is to spend an afternoon building all three at once, and the teams I would bet against are the ones that do exactly that. Five Gems built in a single sitting are five untested instruction sets, none tuned by real use, all producing output that is almost right in ways you have not yet learned to spot.

Build the one covering whatever took most of your time last month, use it for a fortnight, rewrite the instructions once you have seen where the output still needs your hand, and only then start on the next. The first Gem teaches you how specific your instructions have to be for your own work, and that lesson transfers to everything after it.

The post How to build Gemini Gems for recurring crew work appeared first on 麻豆传媒.

Most coordinators never find out that path exists, so a good setup stays where it was made and dies with the person who built it. I want to walk through the whole thing: what a Copilot Notebook actually holds, what to put in one for crew work, how the step from Notebook to agent works, what the approval process involves, and where the ceiling sits for a company that does not have a large IT function.

What a Copilot Notebook is

A Notebook is a container holding the references and working material for an ongoing piece of work, which lets Copilot reason across all of it at once rather than taking one file at a time and losing the connections between them. Ask whether a date change creates a crew conflict and it can consider the schedule, the assignments and the phase sequence together instead of making you paste three documents first.

Two things about availability are worth knowing before you plan any of this around it. Microsoft in June 2026, beyond the Microsoft 365 Copilot users who previously had them, making shared notebook collaboration considerably more available across a team. In July it added Markdown, TXT and RTF files as supported references, with Markdown following in August, which matters for anyone whose site documentation is not sitting in Office formats.

The more eye-catching capabilities need a caveat that vendor-adjacent write-ups tend to skip. When Microsoft announced the April 2026 wave of Copilot updates, which covered referencing SharePoint content and OneNote notebooks, generating Word documents or PowerPoint presentations from notebook content, sharing a notebook with a Microsoft 365 Group, and mind-map views, it described that set as rolling out in preview to users signed up for the Frontier program. Treat those as direction of travel and build this quarter’s work on what your own tenant already has.

What belongs in a crew planning notebook

The pattern is the same one that governs any saved setup: material that changes slower than you will realistically maintain it belongs in the Notebook, and this week’s specific change belongs in the prompt where you can see what you gave it.

From a notebook to an agent the company installs

1. Build and use the Notebook first

Use it for a few weeks before you consider anything downstream, because what you are really doing is finding out whether the references are the right ones and where the output still needs your hand, and that is far cheaper to discover while the thing is still yours alone.

Give it a situation you already handled and compare what it produces against what you actually sent, which tells you three things at once: whether it holds your format, whether it invents anything, and whether it flags the gaps you would have flagged yourself.

Example test prompt: “The GC has pulled the level 3 rough-in on Halston forward by two weeks. Draft the manpower notice in our format, flag any crew this puts in conflict across jobs, and list anything you had to assume.”

Where the output is wrong the fix is almost always a reference or an instruction rather than a cleverer prompt, and corrections made at that level hold for everyone who uses it later rather than needing to be retyped each time.

2. Turn the setup into an agent with Agent Builder

Agent Builder is where a configured setup becomes something with a name and an identity that other people can run without knowing how you built it. The instructions carry the same weight here as anywhere else in this series, and the refusals matter more once other people will be relying on it.

Example agent instructions: “You produce the weekly manpower notice for our mechanical crews across active commercial jobs. Follow the format in the referenced example exactly. Group hours by phase and area. Flag any crew assigned to overlapping jobs at the top of the output, before anything else. Never estimate an hours figure that was not provided. Never state that a certification is current unless it appears in the referenced status list. If a required input is missing, list it as outstanding and continue rather than filling the gap.”

3. Submit it for administrator review

Since April 2026, for administrator review and approval before it is published to the organisation’s Agent Store. Microsoft’s stated purpose is to let organisations scale distribution of high-quality internal agents while keeping IT admin control.

Once approved, the agent appears in the Agent Store under a section called “Built by your org,” where people across the company can find it and install it. The thing a coordinator built in Denver goes through a review and then sits in a catalogue, under a heading saying the company made it, and the Phoenix office installs it like anything else.

Worth knowing what the reviewer is actually looking at, because it shapes how you should write the thing. They see the instructions you wrote, which means the refusals and the terminology are the parts under scrutiny rather than the polish of the output. An agent whose instructions plainly say what it will not do is far easier to approve than one whose instructions read as a general encouragement to be helpful, so writing for the reviewer and writing for quality turn out to be the same exercise.

None of the other major platforms go this far. A Claude Project can be shared with your team, a ChatGPT Project with contributors, a Gemini Gem like a file in Drive, and all three stop at the same place, which is other people using your setup rather than the company publishing it to itself.

4. Answer three questions before you submit anything

All three are considerably cheaper to answer up front than to discover afterwards, and the first one stops more projects than the other two put together.

• Who can approve an agent, and do they know that is now part of their job? Submitting for review requires somebody who owns the Microsoft 365 admin centre and has time to look.
• What does your tenant actually have today, as against what the release notes describe? Preview capability is not a plan.
• Who owns the agent after it is published? An approved agent with no owner ages exactly as badly as an unapproved one.

5. Watch whether anyone uses it

Microsoft added an Agent 365 Dashboard in July 2026, letting leaders track agent usage and adoption across the organisation with drill-down by individual agent. Publishing something and never checking whether it gets installed is how companies end up with a catalogue of agents nobody opens, and the dashboard is the cheapest available answer to that.

Where this falls short

The path assumes an IT function, which is the honest constraint. Plenty of mid-market specialty contractors run on one IT generalist or an outsourced provider, and “submit for administrator review” then means emailing somebody who has never opened Agent Builder and has no reason to prioritise it. That is a reason to find out who that person is early rather than a reason to dismiss the capability.

There is a Microsoft-shaped boundary too, since none of this applies if your company runs on Google Workspace, where the tooling and the ceiling are both different.

What you get without reaching the full path matters more than the ceiling does, because most readers will not reach it this year. A shared Notebook already solves the smaller half of the problem: the setup stops living in one account, a second person can open it, and the thing survives a holiday. That alone justifies the afternoon, and treating the Agent Store as the entry point rather than the ceiling is how people talk themselves out of starting at all.

Approval also does not make the underlying information current, which is the limit worth being clearest about. An agent can pass review, sit in the Agent Store under your company’s name, and be installed by every office, and it will still answer from the documents it was pointed at rather than from the live state of who is on which crew this week. Rigour about how a tool gets published and freshness of the data underneath it are separate problems, which is the difference between an approved assistant reading a document and the kind of agents that work from live workforce data covering team assembly, assignment matching and change digests.

Opinion: approval is the feature, not the friction

The review step reads like bureaucracy and it is the most valuable part of the whole path. A shared setup nobody senior has read is a liability waiting for the wrong week, because private tools do not get checked and unchecked instructions accumulate quiet bad habits.

It also solves a problem specialty contractors have independent of AI. When offices plan workforce in silos they build their tools in silos too, so four offices each spend an afternoon solving the same problem slightly differently and none of the four versions can be improved centrally. A published agent is the version, and improving it improves it everywhere at once.

The sequence that works is the ordinary one that governs what makes a rollout actually stick with any tool: one person who likes it, then a small group, then the company, with review happening at the point where the consequences of getting it wrong justify the step. Most setups never need to reach the Agent Store, and the ones that do will be obvious because people keep asking for them.

The post How to build a Copilot Notebook and turn it into an agent your company can install appeared first on 麻豆传媒.

ChatGPT has two settings that make a bounded version of this workable, and because they do different jobs and live in different places, most people who set up the first one assume they have also handled the second. This walks through what a Project is, how project memory actually separates one general contractor’s information from another’s, where the training setting sits and why it is a separate question, and how to structure projects when you are working for four contractors at once.

What a ChatGPT Project is

A Project is a container that holds related chats, uploaded files and its own instructions together, so work inside it starts with the files already attached and the rules already set.

Three parts matter for crew work: project files, which are documents any chat in that project can reference; project instructions, the rules that apply inside that project; and project memory, which governs what ChatGPT carries between conversations and is the part doing the confidentiality work.

Capacity is generous enough that structure rather than storage becomes the constraint, and OpenAI’s puts file limits at 5 per project on Free, 25 on Go and Plus, and 40 on Edu, Pro, Business and Enterprise, with a maximum of 10 files per simultaneous upload. The number of projects you can create is unlimited, which is what makes the one-project-per-contractor approach below practical.

The two settings people confuse

Almost everything written about AI confidentiality collapses these two controls into one, which is how a coordinator ends up flipping the first, assuming the second is covered, and solving half the problem without knowing it.

That last row is the most useful fact in this piece and it appears almost nowhere else. On a shared project, one person’s setting protects the whole thing.

Setting up projects for multi-contractor work

1. Build one project per general contractor rather than one for everything

The instinct is a single project called “work.” Given that projects are unlimited and memory is bounded per project, the better structure gives each general contractor its own, which keeps their information separate by design rather than by discipline.

Two situations complicate this and are worth deciding before you build. A job with a joint venture or a construction manager between you and the owner is genuinely a job-level project rather than a contractor-level one, and scoping it to the job is the honest answer. Your own internal planning, the roster and certification picture across everything you run, belongs to you rather than any client, so it gets its own project and is the one most likely to hold personal data about your people.

2. Set project memory to project-only

Inside each contractor project, set memory to project-only, after which conversations draw context solely from that project and cannot reach your general chats or anything held in a different project.

If you share the project with a colleague, this happens automatically. OpenAI is explicit that sharing turns project-only memory on and that shared projects have no access to any member’s context or memories from outside the project, which is a sensible default and one most people would not predict.

3. Check the training setting, which is a different place entirely

Open your account settings and find “Improve the model for everyone.” On Business, Enterprise and Edu, OpenAI states that information accessed from projects is not used for training by default. On Free, Plus and Pro it may be used, and whether it is depends on that toggle.

This takes about a minute once you know it exists, and it is the step almost everyone skips because they believe project memory already covered it.

4. Write instructions that name your terminology and your limits

Instructions inside a project let you state once what you would otherwise repeat weekly, and the useful ones combine your vocabulary with explicit refusals.

Example project instructions: “You help me coordinate crews for a mechanical contractor working as a subcontractor to this general contractor. Their manpower report goes out Thursdays in the format in the attached example, hours grouped by phase. A crew is a foreman plus three to five installers moving between jobs as a unit. Never estimate an hours figure I have not given you. Never state that a certification is current unless it appears in the attached list. If information is missing, ask rather than assume.”

The internal planning project needs different instructions, because its job is comparison across contractors rather than work for any one of them.

Example instructions, internal planning project: “You help me see our whole labor picture across every job we are running. Do not draft anything client-facing here. When I ask where we are short, answer by crew and by week, name the jobs competing for the same crew, and state which certifications gate the scopes involved. Never merge this with anything from a client project. If the roster attached here is older than two weeks, say so before answering.”

That last instruction is doing something specific and worth copying. Asking the assistant to flag the age of its own source material will not make the file current, though it does put the staleness question in front of you at the moment you are about to rely on it.

5. Load the files that repeat, and leave out what should not be there

Attach that contractor’s standing job facts and site constraints, the report format they asked for, the phase sequence, the crews assigned to their jobs, and the certifications gating those scopes. Tracking who has what certification and when it lapses belongs in your system of record rather than a chat window, though a project that knows which qualifications a scope requires will stop proposing a crew that cannot legally perform the work.

What stays out is anything whose exposure would be a genuine problem regardless of any setting: personal data beyond what planning requires, anything under a specific confidentiality undertaking, anything you would struggle to explain having uploaded.

6. Test each project before you trust it

Give a new project a situation you already handled and compare its output against what you actually sent. You are checking three things: whether it holds the format, whether it invents anything, and whether it correctly refuses the things you told it to refuse.

Example test prompt: “The level 4 rough-in on the Halston job moved up two weeks. Draft the manpower notice in our Thursday format, flag any crew this puts in conflict, and tell me what you had to assume.”

The refusal test matters most and is the one people skip. Ask it something it should decline, like requesting an hours estimate you never provided, and confirm it says so rather than producing a confident number. An instruction set that has never been tested against the thing it is supposed to prevent has not really been tested.

Where this falls short

A configured boundary is a design intent rather than a guarantee, and it is worth holding that distinction. Project memory is not encryption and it does not constitute a compliance position, so a contractor carrying contractual data-handling obligations to a general contractor needs the right plan tier and the terms that come with it rather than a toggle somebody switched.

The structure also needs maintaining once it exists, because jobs finish and a project full of a completed contract’s material is worth archiving rather than leaving open indefinitely, which means somebody has to own that on a schedule or it will not happen.

The same question is worth asking of every system holding this material rather than only the AI one. The security of the software you rely on gets assessed on its terms, its certifications and its defaults rather than on a setting an individual user remembered to change. A tool that already knows your crews and your jobs should keep each contractor’s picture separate as a property of how it is built, which is one of the things that distinguishes purpose-built AI workforce planning from a general assistant somebody configured carefully on a Tuesday.

Opinion: the prohibition was never going to hold

The standard advice, keep client data out of public AI tools, has the shape of a rule nobody follows. Specialty contractors working across several jobs need help with exactly the material the rule excludes, so what actually happens is that people use the tool anyway, in whatever configuration their account shipped with, and stop thinking about it.

A bounded setup is a better outcome than a prohibition that gets quietly ignored, and two settings plus a project per contractor takes twenty minutes while putting a real boundary where there was previously an unexamined default. It is not a compliance programme and nobody should present it as one, but it is considerably better than the honour system most coordination teams are running on right now.

The post How to set up ChatGPT Projects when you work for several general contractors appeared first on 麻豆传媒.

Projects is the feature that fixes that, and almost nobody in construction has been shown how to set one up properly, so what follows walks through what a Project holds, what to put in one for crew planning, how to share it so the setup survives one person’s holiday, and where the approach stops working. Setting up your first one takes about fifteen minutes.

What a Claude Project is

A Project is a saved workspace inside Claude that keeps three things together: custom instructions that tell Claude how to behave inside that space, knowledge files you upload once that every conversation in the project can draw on, and the conversation history, which stays grouped rather than scattering across your account.

Anthropic’s own documentation on describes them as self-contained workspaces with their own chat histories and knowledge bases, where you upload documents, provide context, and keep focused conversations in one place. Projects are available on every plan including free accounts, and Anthropic also publishes a short walkthrough on if you want the click-by-click version alongside this one.

The practical difference is that a conversation starts somewhere other than zero, so you can ask about the west wing and Claude already knows which job that is, which crew is on it, and that your weekly update follows a particular format because the GC asked for it that way in March.

Project, prompt, or neither: what goes where

Deciding what belongs in a Project and what stays in the prompt is the choice that determines whether the setup helps you or quietly misleads you six weeks from now, and it is worth making deliberately rather than by habit.

The rule underneath the table: information that changes slower than you will realistically maintain it belongs in the Project, and everything else belongs in the prompt where you can see it.

Setting up a crew planning project

There are four things to do and the order matters, though the whole sequence runs to roughly fifteen minutes for your first Project and five for each one after that.

1. Write instructions that constrain, not just describe

Most people write instructions describing what they want and stop there. The instructions that change output quality are the ones that tell Claude what it must not do, because an assistant with no stated limits will fill a gap confidently rather than flag it.

Example instructions: “You help me plan and communicate crew assignments for a mechanical contractor running multiple commercial jobs. A crew means a foreman plus three to five installers who move between jobs as a single unit rather than being assigned one name at a time. Always report hours by phase and area. Never estimate an hours figure I have not given you. If a certification or licence is missing for a scope, say so plainly rather than assuming it is current. If you had to assume anything to answer, list the assumptions at the end.”

The last three sentences are doing most of the work. Everything before them is context; those are the guardrails that make the output checkable rather than merely plausible.

The difference between a vague instruction and a specific one shows up quickly when you write both out and compare what each one actually asks Claude to do.

Too vague: “Be accurate and helpful when discussing crew scheduling. Use professional language.”

Specific enough to change the output: “Report hours by phase and area. If a crew is assigned to two jobs in the same week, say so before answering anything else. Never estimate an hours figure I have not given you.”

The first version cannot be wrong, which is precisely why it does nothing, whereas the second tells Claude what to check and what to refuse, so the output either meets the standard or visibly fails to.

2. Load the knowledge files that repeat

Upload the material from the “Project knowledge” rows above. Five well-chosen documents beat forty, because a project holding everything produces vaguer answers as the useful signal gets diluted by material irrelevant to the question.

For a crew planning project that comes down to the current roster with crew composition, the certification and expiry list, a real look-ahead you have already sent so the format is demonstrated rather than described, the phase sequence for active jobs, and the labor curve from the estimate.

One caution before you upload a roster: personal data beyond what the planning actually requires should stay out, and it is worth understanding where your data actually goes before rather than after.

3. Test it against a week you already handled

Before trusting a new Project, give it a change you dealt with last month and compare what it produces against what you actually sent. You are checking whether it holds your format, whether it invents anything, and whether it flags the gaps you would have flagged.

Example test prompt: “The GC on the Riverside job has pulled the level 3 rough-in forward by two weeks. Draft the manpower notice in our usual format, flag any crew this creates a conflict for, and list anything you needed to assume.”

Where the output is wrong, the fix is nearly always an instruction rather than a better prompt, so add the correction to the instruction set and it holds for every conversation afterwards instead of needing to be retyped.

Run the refusal test as well, since it is the one people skip: ask for something the instructions say it should decline, such as an hours figure you never supplied, and confirm it says so rather than producing a confident number.

Example refusal test: “How many hours should we carry for the level 3 ductwork next month?”

If it answers with a figure rather than telling you it does not have one, the instruction did not take, and that is worth knowing before the Project is drafting anything that leaves the building.

4. Share it so it outlives you

On Team and Enterprise plans a Project can be shared with permissions controlling who can change what. Several people then work against the same instructions and the same knowledge base, which is what stops the setup living in one person’s head.

The practical payoff is ordinary and worth stating: a second coordinator covering a holiday does not start from a blank window, and someone new to the role is useful in days rather than after a month of asking colleagues how things are done here. If you are building something you expect the team to rely on, share it early, while the instructions are still short enough for someone else to read.

Where this falls short

A Project’s knowledge files are a snapshot taken on the day you attached them, and nothing tells you when the snapshot has gone stale. Upload a roster in March, run the Project through August, and it will answer with total confidence using the March roster, including the two people who left in May and the certification that lapsed in June. An assistant that says it does not know is a minor irritation, whereas one that produces a clean, well-formatted, plausible answer built on stale data is worse than not asking, because nothing in the output signals that it needs checking.

That is the same instinct the industry already has about AI, and it is correct: garbage in, garbage out is the most repeated phrase in any practitioner conversation about this technology. A Project does not escape the problem so much as relocate it, from the data in your systems to the files somebody remembered to re-upload.

There is a plainer risk alongside it. A Project is a separate window living outside the tools you already have open, and separate windows are the things people stop opening around week three. If it does not get into the Monday routine within a fortnight, it will not survive the month.

The structural answer is to stop maintaining a copy of information that already exists somewhere current, which is what a connection layer is for, with the systems where your real work lives becoming things an assistant reads directly rather than things you export and upload. 麻豆传媒 ships an MCP connection for this reason, which is how purpose-built AI workforce planning differs from a general assistant you have configured carefully: one holds the current plan, the other holds a copy of it.

Opinion: the setup matters more than the model

Every few months a new model arrives and the conversation restarts around which one is best. For the work described here, that question matters far less than whether you have spent fifteen minutes on the setup, and I would rather hand a coordinator a well-configured Project on a year-old model than a blank window on the newest one.

The reason is that construction’s AI problem has never really been capability. Contractors who plan crews across multiple jobs carry an unusual amount of standing context, and the tools have been perfectly able to use it for a while now, but there has been nowhere to put it. If you only do one thing from this piece, write the instruction set. The knowledge files can wait a week and the sharing can wait a month, whereas the instructions are what change the output tomorrow morning.

The post How to set up Claude Projects for crew planning appeared first on 麻豆传媒.

Already trusted by 40% of top contractors, 麻豆传媒 is the first to launch agentic AI that helps construction better manage one of its scarcest resources: its people.

TORONTO 鈥� July 29, 2026 鈥� 麻豆传媒, the AI company redefining how construction plans its workforce, today announced the latest expansion of 麻豆传媒 AI, introducing 麻豆传媒 agents and the 麻豆传媒 MCP. Together, they extend the 麻豆传媒 AI platform from answering workforce questions to taking action on planners鈥� behalf 鈥� and bringing live 麻豆传媒 data into the AI tools contractors already use.

AI is reshaping much of construction by reading contracts, flagging safety risks, and sharpening estimates. But most of it has pointed at project execution, not workforce management 鈥� despite the fact that today鈥檚 difficult labor market makes staffing decisions one of the biggest levers on profitability and retention. 麻豆传媒鈥檚 AI is the first and only tool to address that gap.

From Answers to Action

麻豆传媒 introduced Ask 麻豆传媒 earlier this year, letting construction teams ask questions about their data in plain language: who鈥檚 available, when, and for how long. Agents build upon that foundation and go from answering to doing. 麻豆传媒 agents can execute workforce planning tasks such as assembling an optimized project team, identifying and resolving scheduling conflicts and resource constraints, creating customized reports, and recommending the best person to fill a gap.

鈥溌槎勾� AI started by helping teams get answers faster. Agents are the next step 鈥� they get the work done,鈥� said Mallorie Brodie, CEO and Co-Founder of 麻豆传媒. 鈥淲orkforce planning is a complex process, and our AI makes it easy to sift through complex information, recommend action, and streamline high-effort, high-friction workflows with agents that help contractors keep their business moving.鈥�

Grounded in the Industry鈥檚 Deepest Workforce Data

麻豆传媒 understands the complexity of workforce planning better than anybody else, and its agents are built on the same deep, structured workforce data that has made 麻豆传媒 the planning platform of choice for 40% of top contractors. When an agent recommends a team or flags a conflict, the reasoning is grounded in a company鈥檚 own project and people data inside 麻豆传媒 鈥� so recommendations reflect how each unique business actually works, and sharpen over time as more of that data accumulates.

Five Agents, Built for the Work That Takes the Most Time

麻豆传媒 AI launches with five agents, each targeting a distinct planning task:

• Team Builder proposes strong project teams based on a combination of your firm鈥檚 unique criteria and historical workforce data. Operations leaders review, adjust, and confirm 鈥� building on a solid draft rather than filling roles in isolation.
• Assignment Finder helps operations leaders quickly identify suitable roles for people with upcoming availability, keeping workers fully utilized and off the bench.
• Project Change Digest summarizes what changed on a project in a structured digest, so leaders and executives stay current without digging through audit logs.
• Project Org Chart builds visual org charts that convey hierarchy and reporting lines for pursuit bid packages and active project planning.
• Workforce Brief builds a lookahead schedule to provide visibility of upcoming changes, including assignments starting or ending and planned time off.

When 麻豆传媒 agents suggest an action, they always provide the reasoning behind it, allowing the user to review and decide what to implement. Every step is easy to edit or override before it takes effect.

“That鈥檚 not AI magic; it鈥檚 purposeful design at work,” said Vincent Seguin, Chief Technology Officer of 麻豆传媒. “Every agent we build has to show the recommendation and the reasoning behind it, in plain language, before a planner ever sees it. If we can’t explain why an agent did something, it doesn’t ship. That’s the bar for building AI that earns a team鈥檚 trust and truly makes their jobs easier”

麻豆传媒 Data, Inside the AI Tools You Already Use

Alongside agents, 麻豆传媒 is also launching the 麻豆传媒 MCP (Model Context Protocol) 鈥� a secure connection that brings live 麻豆传媒 data into the AI assistants contractors are already building workflows around, including Claude, ChatGPT, Copilot, and Gemini. A contractor can ask their AI assistant a workforce question in plain English and get a response powered by live data from their 麻豆传媒 account 鈥� without switching tools or pulling a report. Combined with data from a CRM or timesheet tool, planners can answer questions that no single system could address on its own.

鈥淭eams are already incorporating AI assistants into their workflows,鈥� said Brodie. 鈥淲ithout an MCP, those tools aren鈥檛 tapping into their 麻豆传媒 data, which represents some of the most valuable workforce data they can leverage for planning. 鈥�

Availability

麻豆传媒 agents are available now as part of 麻豆传媒 AI. Interested contractors can learn more and request a demonstration at gobridgit.com.

About 麻豆传媒

麻豆传媒 is the AI company redefining how construction plans its workforce. Trusted by 40% of top contractors, 麻豆传媒 gets the right people on the right projects by bringing people and projects into one clear view, surfacing the insights that improve every staffing decision, and acting on them to reduce manual, high-friction tasks. Contractors get stronger project teams, smarter staffing and bidding decisions, and a workforce strategy that stays ahead of demand instead of reacting to it. 麻豆传媒 is a privately held company backed by investors such as Autodesk, Salesforce Ventures, and Sands Capital, among others.

For more information, visit gobridgit.com.

Media Contact:
Amy Palmer, Vice President of Marketing
麻豆传媒
amy.palmer@gobridgit.com

The post 麻豆传媒 Launches the First AI Agents Built for Workforce Planning appeared first on 麻豆传媒.

That matters because the documents are where the hours go. 61% of construction firms now use AI or plan to increase their investment, and the teams getting real value tend to point it at the reading-and-writing load first. This guide walks through where Claude earns its place on that load, how to set it up so you’re not starting from scratch every time, and the lines worth holding.

Where Claude is strongest: long documents

The standout use is feeding Claude a long document and asking real questions about it. Upload a full spec section and ask where the submittal requirements for a particular division live. Paste a subcontract and ask it to pull out every clause that touches schedule or liquidated damages. Drop in a meeting transcript and ask for the decisions and who owns them. Work that used to mean an afternoon of scrolling and highlighting becomes a few minutes of asking and checking.

The reason this works is that Claude can hold a lot of text in front of it at once. You don’t have to feed a contract to it in pieces and hope it remembers the early sections by the time it reaches the end. You can give it the whole thing and ask questions that span the document, which is the kind of reading that takes a person real time and concentration. For a coordinator staring at a stack of submittals or a PM trying to find the one provision that changes how a change order gets priced, that is the difference between a quick answer and a lost afternoon. The payoff is not unique to construction: McKinsey finds frontline teams gaining when AI takes the repetitive load off their plate.

A concrete version comes up before almost every bid: you’re handed a geotechnical report and a structural spec the week before the date, and you need to know what they say about deep foundations and who carries the risk if conditions differ. Rather than read both cover to cover, you load them and ask. Claude points you to the relevant sections and summarizes them, and then you read those sections closely to confirm. The reading you do is the reading that matters, and you skip the part where you hunt for it.

A habit that pays off with long documents is to work in focused steps instead of one sweeping question. Ask what the spec says about a division, then narrow to the clause you actually need, then ask how it lines up with the submittal. When you already know exactly what you are after, a single detailed prompt is fine, but while you are still exploring, the step-by-step path keeps the answers grounded and easy to check against the source.

This is also where Claude and a quick chat tool part ways. If most of your AI use is short documentation tasks, a companion guide on how construction project managers use ChatGPT covers that ground well. Claude shines when the input is long and the question requires reading across all of it.

Setting up Claude for the work you repeat

Most of a construction team’s writing is the same handful of jobs done over and over: the weekly owner update, the daily report, the standard coordination note, the format your minutes always take. Claude lets you set that up once instead of re-explaining it every time, through a feature called Projects.

A Project is a saved space where you load the context you reuse: your reporting template, a style you want matched, the standing facts about a job, the documents you keep referring back to. Once it’s there, every conversation in that Project already knows it. Your Tuesday owner update stops being a blank page and starts being a five-minute pass over a draft that already follows your format. The same goes for a recurring submittal review or a standard RFI response.

The payoff compounds the more you use it: a prompt you refine over a few weeks, sitting inside a Project that holds your templates and standards, becomes a piece of your actual workflow instead of something you experiment with on the side. A regional team might keep one Project for owner reporting and another for subcontract review, each loaded with the formats and standing facts that job needs. The habit of building reusable setups is worth developing early, and it carries directly into any AI tool you adopt later.

Drafting that follows your instructions

Claude is careful about following detailed instructions, which makes it useful for drafting where the format and tone actually matter. Give it the situation and the constraints, and it holds to them: an owner update in the structure your client expects, a change-order narrative that explains the cause without editorializing, an RFI written to the point, a coordination note to a sub that lands the right level of firmness.

The practical move is to be specific about what you want. Tell it who the reader is, what they already know, what you need from them, and how long it should be. The more you treat it like briefing a capable assistant rather than typing a search query, the closer the first draft lands to something you can send after a quick edit. You stay the author; Claude just clears the blank page.

It helps to give it raw material to work from. Paste the thread you’re replying to, the clause you’re enforcing, or the bullet points you’d have scribbled on a notepad, and ask for the version you’d actually send. A change-order narrative built from the daily reports and the RFI that triggered it comes back in minutes, in the structure your owner is used to seeing, ready for you to correct the one detail only you would catch.

Comparing and checking documents

Because Claude works across long documents, it can do a useful first pass at comparing them. Give it two versions of a submittal and ask what changed. Put a spec section next to a submittal and ask where the submittal might not line up. Hand it last month’s schedule narrative and this month’s and ask what shifted. It surfaces the spots worth a closer look faster than reading both end to end. On a busy submittal log, that first pass is often the difference between catching a missing certification this week and catching it after it has already held up a delivery.

The honest framing matters: what you get is a list of candidates to check by hand, and the reviewer is still you. Claude can flag that a submittal seems to miss a required certification; it cannot be the final word on whether the submittal conforms. Treat the output as a faster way to find the differences that deserve your attention, then make the call yourself. Used that way, it shortens the search while leaving the judgment where it belongs.

Where Claude still falls short

Start with the gap every general assistant has: Claude works only with the text you give it, and it has no access to your project records. It has not seen your job histories, your live schedule, or which of your people have run hospitals, so a staffing question comes back as generic advice dressed up as a specific answer. The same shortfall scales to the company level, where 85% of AI projects fail on data quality, because the model can only reason over records you can actually hand it, and most contractors’ records sit in systems that do not talk to each other.

Claude’s comfort with long documents has a flip side worth naming. Holding a whole contract in view is not the same as reading every line the way you would, so it can still misread a requirement, point to the wrong code section with total confidence, or paraphrase a clause in a way that quietly drops the part that governs. One construction attorney told to see a general tool drafting a construction contract. The discipline that follows is to let Claude locate and draft, then confirm anything carrying code, contract, or safety consequences against the original before you rely on it.

Some calls sit outside any general tool’s reach no matter how well it handles text. Deciding who runs your next project is not a writing task; it turns on build history, the relationships a person carries, and how far they will be driving every morning. That is the work 麻豆传媒’s purpose-built AI workforce planning is built for, reasoning over your own verified people and project records while a person makes the actual call. Claude earns its place as a reading and writing partner; staffing decisions belong on a system built to weigh the workforce data behind them.

Getting real value from Claude

Used well, Claude is the team’s reader and drafter. It takes on the long documents and the repeatable writing, and it leaves the judgment with the people who carry the project in their heads. The teams who get the most out of it build Projects for the jobs they repeat and keep a human signing off on anything that counts.

A good first step is to go narrow on purpose: take one document-heavy job you already do every week, like turning a coordination transcript into minutes or pulling the key clauses out of a subcontract, and build a Project around it. Get that one reliable and the next few will suggest themselves, and you will have learned where Claude genuinely helps and where your own read still has to lead. If crew planning is that job, the walkthrough on setting up a Claude Project for crew planning takes it step by step.

The post How construction teams use Anthropic鈥檚 Claude appeared first on 麻豆传媒.

The answer, used well, is a real dent in the documentation load that eats every week. Across the industry, 61% of construction firms now use AI or are putting more behind it, and for Google Workspace shops, Gemini is often the first place that money shows up, because nobody has to install anything. This guide covers where it helps in the flow of your existing tools, the one feature that does the most to earn a skeptic’s trust, and the limits worth respecting.

Where Gemini fits: inside Google Workspace

The simplest way to think about Gemini is the AI built into the Google tools you already open. In Gmail, it will draft a reply to a sub from a one-line instruction about what you need. In Docs, it will summarize a long document or tighten a paragraph. In Meet, it can take notes and produce a summary of who said they would do what. None of that asks you to change where you work; it adds a helper to the place you were already going to be.

That matters more than it sounds for adoption. A general contractor’s operations team does not have spare hours to learn a new platform, and most people quietly avoid tools that live outside their routine. When the assistant is already in the inbox and the document, the cost of trying it is close to zero. You ask it to boil down the thread you are staring at, see whether the result is any good, and decide from there. The tools meet people in the flow they are already in, which is usually where adoption actually sticks. The teams that do adopt tend to pull ahead, with PwC reporting since 2022.

The same logic applies to meetings, where the payoff is concrete. A coordination call or an OAC meeting that used to mean a half hour of writing afterward can come back as a draft summary with action items, generated from the Meet recording. You still read it for accuracy and nuance, but the blank-page part of the job is gone, and the notes get written even on the weeks when someone would otherwise have skipped them. On a project running several coordination meetings a week, that is hours returned across the month and a record that no longer depends on whoever had time to type it up.

Answers grounded in your own documents

The feature most likely to win over a skeptic is NotebookLM, a Google tool that answers questions using only the sources you give it. You upload the documents that matter for a job, a spec set, a batch of RFIs, a contract, a stack of submittals, and then ask questions that get answered from those documents and nowhere else. Crucially, it shows you where each answer came from, citing the passage in the source so you can click straight to it and confirm.

For a construction audience, that citation behavior is the thing worth paying attention to. The deepest worry about AI on a jobsite is that it makes things up, and a tool that quotes the source passage behind every answer addresses that worry head on. Ask what the submittal requirements are for a division and it points you at the lines in the spec that say so. Ask which RFIs touched a particular detail and it shows you the ones it pulled from. You are not trusting a black box; you are reading the source it found, faster than you would have found it yourself.

The honest caveat is that grounding reduces the risk, it does not erase it. The tool can still pull the wrong passage or miss one, so the discipline is the same as with any source: read the citation it gives you before you act on the answer. Used that way, it turns “find me where the contract addresses delay” from an afternoon into a couple of minutes, with the receipts attached. That habit of getting the information into one trustworthy place is the same groundwork behind generally, and it is what makes any of these tools worth using.

Drafting and summarizing in the flow

Most of the writing a construction team does is the same set of jobs on repeat, and Gemini handles them where the work already lives. An owner update gets drafted in the Doc you were going to write it in. A reply to a sub about a schedule change gets drafted in the Gmail thread you were already reading. A weekly progress summary gets pulled together from the notes you keep in Drive. The work is the documentation overhead that does not require a PM’s judgment, just their time, and that is exactly the kind of task a writing assistant is good at.

The move that makes the output usable is being specific. Tell it who the reader is, what they already know, and what you need from them, and the draft comes back close enough that a short edit gets it out the door. A note to a sub that you would have put off until the end of the day becomes a two-minute task, which is often the difference between coordination that happens on time and coordination that slips. You do not need to type the request perfectly to get there. Typos and half-finished sentences are fine, and Gemini reads plain references the way a colleague would, so “the airport” or “the Seattle job” lands without you spelling out every detail. If your day is more about short, standalone documentation tasks than living inside Google’s tools, the companion guide on how construction project managers use ChatGPT walks through the same kinds of jobs in a standalone assistant. The principle holds across all of them: you brief it like an assistant, you stay the editor.

Searching across your Drive

A project’s files tend to sprawl across a Drive, and finding the current version of the right document is its own small tax on the day. Gemini can search across your Drive in plain language, so instead of remembering which folder holds the latest site logistics plan, you ask for it. It can also pull together what several documents say on a topic, which helps when the answer you need is spread across a few files nobody has consolidated.

This is genuinely useful and also where a familiar limit starts to show. The assistant searches what is actually in your Drive, organized the way your team organized it. If three versions of a plan live in three folders with three naming conventions, Gemini will work with that mess rather than fix it, and the cleaner your shared files, the better the answers. The tool rewards teams that keep their documents in order, which is worth knowing before you judge it on a chaotic Drive.

Where Gemini still falls short

Begin with what Gemini cannot do, because it is the same ceiling every general assistant hits: it does not hold your structured project or workforce data. It can read what sits in your Drive, but it has no picture of who has built data centers, who frees up in the fall, or who has a history with this owner. A staffing question gets answered from broad patterns, which is the wrong footing for a decision that rides on your own records.

It can also be confidently wrong. Step outside NotebookLM’s grounded answers and Gemini will sometimes state a code requirement that is off, gloss a provision so the governing clause disappears, or turn a genuine ambiguity into a clean sentence that simply reads true. Treat its drafting and searching as fast first passes, and confirm anything with code, compliance, or safety on the line before you rely on it. One boundary is worth saying out loud as well: Gemini lives in Google Workspace, so a company standardized on Microsoft 365 will find its in-the-flow assistant is Copilot instead, with the same habits carrying over.

The highest-stakes calls stay out of reach for any general tool. Choosing who runs your next hospital is not a drafting problem; it depends on build history, the relationships a person brings, and the commute that quietly decides whether they stay. Weighing all of that is what 麻豆传媒’s purpose-built AI workforce planning is for, reading your own people and project history while the decision stays with a person. Gemini is built to help you write and find; the staffing call wants a tool that weighs your workforce.

Getting real value from Gemini

What makes Gemini stick is how little it asks of you to start. It is already in the inbox and the document, so trying it is just letting it handle the next summary or draft. Use it where your Google work already happens, reach for NotebookLM when getting the answer exactly right is the point, and keep a person on every call that carries weight.

If you want a first move, pick one thing that already runs through Google, like turning a Meet recording into minutes or loading a job’s documents into NotebookLM and asking your questions there. It is a low-risk place to begin, and a couple of weeks in you will know which corners of your week Gemini earns and which still need your own read. For the work that comes back every week, building a Gem for recurring crew work is the next step up from one-off prompts.

The post How construction teams use Google鈥檚 Gemini appeared first on 麻豆传媒.

This guide walks through the ones that matter: where your data actually goes, how access should work, how to keep from acting on a wrong answer, and where to keep a person firmly in charge. None of it is about trusting the technology more than it deserves; it is about setting the conditions under which trusting it is reasonable. That is the same judgment a construction leader already applies to a new subcontractor, a piece of rented equipment, or a first-year hire: define where it can operate, check its work, and keep the high-stakes calls with someone accountable. Get these right and you can let your team use these tools without lying awake about it.

Know where your data goes

The first question is usually the sharpest: if we paste a bid strategy or project financials into an AI tool, does that become training data for a system the whole world can use? On the business versions of the major assistants, the answer is no, and the vendors put it in writing. Microsoft, for example, states that with Copilot your prompts, inputs, and responses are . The other major business assistants make similar commitments for their paid and enterprise tiers. Construction teams put exactly this question to 麻豆传媒 in a , and the answer was plain: your data is your data, and it stays in your account. In practical terms, the cost spreadsheet you paste in to clean up, or the owner email you ask it to soften, stays inside that account and is not fed into a system a competitor might later query. For a business where the margin on a pursuit can hinge on what nobody else knows, that boundary is the whole ballgame.

The practical catch worth knowing is that the tier matters. Free, consumer versions of some tools handle data differently than the business and enterprise plans, and the difference is exactly the kind of thing that should be settled before sensitive information goes anywhere near a prompt. The move for a construction company is straightforward: use the business or enterprise version your company controls, and confirm the data policy in writing rather than assuming it. Ask the vendor directly what happens to your inputs, where they are processed, and whether anything you type trains a model used by other customers. A good tool answers that question cleanly.

AI should work within the permissions you already have

The second worry is internal: if an assistant can read across company files, can someone use it to see cost data, salaries, or documents that are not theirs to open? The right answer is that a good business tool respects the access controls you already have. Microsoft 365 Copilot, for instance, inherits your existing permissions and sensitivity labels, so it will not show a person a document they could not already open on their own. The AI does not become a backdoor around the rules your IT team set.

This is worth making explicit when you evaluate any tool, because it is not automatic across every product on the market. The question to put to a vendor is simple: does the assistant honor our existing permissions, or does it create a new way to reach data? If the answer is anything other than a clear yes on honoring your controls, that is a reason to slow down. The tools worth adopting treat your permission structure as a boundary to respect. In practice that means the same wall keeping a field engineer out of executive cost data also keeps the assistant from handing it over, however the question gets phrased, which is exactly the behavior you want confirmed before you turn a tool loose across a company’s files.

Keeping a wrong answer from becoming a decision

The risk that makes experienced people nervous is that AI states things with total confidence whether or not they are true. It will cite a code section that does not exist, summarize a provision in a way that misses the clause that governs, or invent a plausible answer when it does not actually know. This is real and widespread, with a Qlik survey finding , the soil confident-but-wrong answers grow in. It is also manageable with a few habits.

Verify anything that carries weight against the source. Use the AI to find the relevant spec section or draft the language, then read the original before you act on it. Where you can, ground the tool in your own documents so its answers come from your material rather than the open internet, and check the citation it gives you. Be specific in what you ask, and break a big question into smaller steps when you are exploring instead of firing one sweeping prompt and trusting the result. When the answer points to a spec clause, open the clause; when it quotes a number, trace it. That check takes a minute, and it is the difference between catching a wrong code reference at your desk and catching it in the field after it has already cost you. These are the same habits that separate the companies getting value from the ones whose pilots fail, since 85% of AI projects fail on data quality, and confident-but-wrong output is a symptom of the same problem. One construction attorney told Construction Dive she to hear of anyone using a general tool to generate a contract, and she is right to draw the line there: the tool can draft the document, but a person decides what it should say.

Keep a person in the loop

The line that matters most is the one between a tool that suggests and a tool that acts. The safe posture, and the one the better AI products are built around, is that the assistant proposes and a person approves. It can surface the candidate, draft the email, flag the expiring certification, and assemble the report, but a human makes the call on anything that carries consequence. You stay the decision-maker; the AI gets you to the decision faster.

That principle holds even as these tools start to take multi-step actions on your behalf. The version worth adopting still stops and waits for your approval before anything happens that affects a schedule, a budget, or a person’s assignment. Asked in that same webinar whether the tool would ever start making changes on its own, the answer was a flat no: it offers options and shows its reasoning, and the person keeps the final say. When you evaluate a tool that promises to do more than answer, the question is whether it keeps a person in the approval seat for decisions that matter. If it tries to act on its own where the consequences are real, treat that as a liability rather than a selling point. The same logic that stops a junior from committing the company on a handshake applies to software: the more a tool can do, the more it matters that a person signs off before it does it.

Where to draw the line

It helps to sort AI tasks into three buckets, because the safe answer is different for each. Treat the table below as a starting point; your own risk tolerance will move a few items around.

The pattern across the table is consistent: the further a task moves from “turn this text into cleaner text” toward “make a judgment that carries risk,” the more a person needs to own it. Most items have a safe version and an unsafe version of the same task. Using AI to find the delay clause in a contract sits in the first column; using it to decide whether a delay claim will hold sits in the third. Keep that distinction in mind and AI takes work off your plate without taking on risk you cannot afford.

Your data foundation is the real safeguard

Underneath every one of these habits is a single truth: AI is only as safe and useful as the data beneath it. A tool grounded in clean, organized, trustworthy information gives answers you can act on; a tool pointed at scattered, stale, conflicting records gives confident nonsense, no matter how careful your prompts are. Getting your information into one place you trust is the groundwork that makes everything else work, which is the subject of .

This is also why the highest-stakes decisions stay outside a general tool’s reach. An assistant can draft a staffing plan, but deciding which superintendent should run your next hospital depends on build-type experience, client relationships, and commute, the kind of structured workforce data a general assistant does not have. That is the thinking behind 麻豆传媒’s purpose-built AI workforce planning: AI that works on top of your own verified people and project records, with a person making the final call. Safe AI and good data are the same project. Build the foundation, keep a person in the loop, and verify what carries weight. Do that and these tools become something your team uses with confidence, and something leadership can finally stop bracing against.

The post Using AI safely on construction projects appeared first on 麻豆传媒.

The free version, Copilot Chat, comes with most Microsoft 365 business plans. It is a chat assistant you can open inside the apps, and it works from the open web plus whatever files you hand it directly. It does not reach into your company’s wider data on its own. The paid version, Microsoft 365 Copilot, is an add-on license that is : Word, Excel, PowerPoint, Outlook, Teams, and SharePoint. It can work across the documents, emails, and meetings you already have permission to see. Both are useful. The paid one earns the “assistant who knows your work” description, because it can actually reach your work. A practical way to tell which you have: if Copilot can answer questions about your own emails and SharePoint files, you are on the paid Microsoft 365 Copilot; if it mostly helps with the document in front of you and the open web, that is Copilot Chat, and your IT or operations lead will know which licenses the company bought.

That distinction matters for a construction team deciding where to start, especially since 61% of construction firms now use AI or expect to before long, and a lot of that spending is simply turning on tools the company already pays for. This guide covers what Copilot does across the Microsoft apps you live in, what happens to your data, and where to stop.

Where Copilot fits: inside the Microsoft tools you already use

The appeal of Copilot for a Microsoft shop is the same as the appeal of any in-the-flow assistant: it shows up where you already work instead of asking you to go somewhere new. In Word it drafts and rewrites; in Excel it explains a dataset and suggests the formula; in PowerPoint it builds a deck from a document and holds to your company template; in Outlook and Teams it handles the communication load covered below. In practice that means a coordinator can drop a labor or cost spreadsheet into Excel and ask why a number moved or which figures look off, and get an explanation in plain language instead of building the pivot table first. The reason this drives adoption is that there is almost nothing to learn. The button is already in the ribbon of the app you opened anyway.

Copilot can now also take multi-step actions inside those apps, which Microsoft calls . Instead of one instruction at a time, you can ask it to reorganize a document, insert a summary, and fix the formatting as a single supervised sequence. The word to hold onto there is supervised. It proposes and executes the steps, and you stay the one who reviews and approves the result, which is exactly the posture you want for anything that ends up in front of an owner or a sub. Used on a long report or a deck, that means you ask for the whole cleanup in one request and then read the result closely before it goes anywhere near an owner.

Meetings in Teams and threads in Outlook

The highest-return starting point for most teams is the communication overhead, and this is where Copilot is strongest day to day. In Teams, it can produce a meeting recap from the transcript: what you missed if you joined late, the decisions that got made, and detailed notes with action items and who owns them. For a team running coordination calls and OAC meetings across several jobs a week, that turns a half hour of post-meeting writing into a few minutes of review, and the notes get captured even on the weeks someone would have let them slide. The recap can go to the people who could not make the call, so the PM who was on a site walk during the OAC meeting still gets the decisions and their action items without chasing anyone for them.

Outlook gets the same treatment for email. Copilot can summarize a long thread so you can catch up on a dispute without scrolling through forty replies, and it can draft a response in the tone you ask for, whether that is a firm note to a sub or a measured update to the owner. With the paid version, it can also pull in relevant context from your calendar and related messages, so the draft already reflects what is actually on your schedule. As with the standalone assistants covered elsewhere in this series, like the guide to how construction project managers use ChatGPT, you stay the editor and Copilot gets you past the blank reply.

Working from your own company files

Because Microsoft 365 Copilot connects to the files in your SharePoint and OneDrive, it can answer questions using your own documents rather than the open internet. Ask what your standard subcontract says about delay, or pull the key points from a folder of project documents, and it works from the material your team has actually stored, limited to what you personally have permission to open. For a team that keeps its specs, RFIs, and standards in SharePoint, the answer to a question like what the standard insurance clause requires comes back in seconds with the source attached, drawn from documents the company already maintains.

That permission boundary is worth understanding, because it cuts both ways. It is a safeguard, since Copilot will not surface a document to someone who could not already open it themselves. It is also a dependency, because the quality of the answers tracks the quality and organization of what is in your SharePoint. A tidy, well-structured document library gives useful answers; a sprawl of duplicates and inconsistent folders gives muddy ones. The tool works with the house you have built, so the cleaner that house, the more the assistant is worth.

Specificity is what turns a vague answer into a useful one. Ask ten people a loose question and you get ten different answers, and an assistant is no different, so name the exact project, the role, or the document you mean and the response narrows to what you were actually after. The teams who get the most out of Copilot tend to be the ones who know their own files well enough to point it straight at the right one.

Your data stays in your tenant

The question that stops most construction firms from trying any of this is what happens to sensitive company data, and on Microsoft 365 the answer is specific and worth knowing. Microsoft states plainly that with Copilot, your prompts, inputs, and responses are . Your bid strategy and your project financials are not becoming training data for a public system, which is the specific fear that keeps a lot of construction leaders from trying anything at all.

Copilot also , sensitivity labels, and retention policies, so it operates inside the access rules your IT team already set. Someone without permission to see cost data does not get it by asking Copilot for it. For a leadership team weighing whether AI means handing the keys to an outside system, that is the reassurance that matters: the work stays inside the environment you already control, under the rules you already wrote.

Where Copilot still falls short

Set the limits next, because they decide where Copilot is safe to lean on. The first is the ceiling every general assistant shares: it has no grasp of your structured project or workforce data. It can read the documents in your tenant, but it does not know who has built hospitals, who opens up in the fall, or who has history with this owner, so a staffing question comes back as broad advice rather than an answer grounded in your operation.

It can be confidently wrong where it matters most. Copilot will occasionally assert a code requirement that is off, return a tidy fire-rating answer aimed at the wrong section, or compress a provision until the controlling clause is gone. The working rule is the one that holds for every tool in this series: let it find and draft, and check anything that carries code, contract, or safety weight against the source. A scope note belongs here too, since Copilot is the Microsoft 365 assistant, so a Google Workspace shop will be using Gemini for the same jobs with the habits unchanged.

Some decisions stay beyond any general assistant, however well it reads your files. Picking the superintendent for your next hospital is not a documents question; it rests on build history, the relationships that person carries, and a commute long enough to cost you the hire. That weighing is the job 麻豆传媒’s purpose-built AI workforce planning was built for, working from your verified people and project records while you keep the final call. Copilot is strong with language and documents; the staffing decision wants a system built around your workforce.

Getting real value from Copilot

Copilot pays off fastest when a team aims it at the communication load first, the Teams recaps and the Outlook threads, then leans on its reach into SharePoint when an answer has to be right, and keeps a person deciding anything with real consequence behind it. Since it already sits in the apps the company pays for, getting started is mostly a decision to begin. It helps to be realistic about the horizon, too, since Deloitte finds most AI investments take , which is a good reason to bank the easy wins early.

A sensible first move is one routine that already runs through Microsoft 365, like turning each week’s Teams coordination meeting into clean minutes with owners and due dates. Prove it there, and you will quickly see which parts of the week Copilot should own and which still belong to your own judgment. From there, building a Copilot Notebook into an installable agent is how that routine becomes something the whole company can run.

The post How construction teams use Microsoft鈥檚 Copilot appeared first on 麻豆传媒.