A company-wide AI usage surveyDesigned, run, and summarized — the first honest picture of how AI is actually used here, the field included.
AI Policy Task Force
Help decide how we build with AI.
Leadership has approved a formal Company AI Policy. You've been asked to help write it — not review it after the fact. This is the real thing.
A three-minute read, then a short intake at the end.
The mission
What we're actually doing
Setting the direction for how Collage uses AI — deliberately, with guardrails people can actually follow.
AI is already part of the work — in the software Collage licenses, on people's phones, in tools they reach for on their own. The question was never whether Collage uses it. It's how: what we adopt, where the guardrails go, and how we protect our people, our clients, and our work while getting real value out of it.
One principle runs under all of it: AI does not transfer responsibility. It doesn't excuse an error, and it isn't an authority on code, law, or a contract. It's a tool — and anything it helps produce is held to the same standard of care as any other work that carries the Collage name.
Why now
This isn't hypothetical.
The question isn't whether our people use AI. It's whether we can see it.
A review of four active Collage contracts found data-handling obligations on every one of them — and not one uses the words "artificial intelligence." The rules that restrict AI are the confidentiality and data-handling obligations that were already there.
12 of 30
office staff using AI today — self-reported, and almost certainly an undercount
Unmeasured
field use — superintendents and PMs outside the office are entirely unseen
Zero
managed accounts, policy, or controls — every use today runs on a personal account
Prohibition doesn't reduce AI use — it relocates it. A ban just moves this work onto personal phones and personal accounts, where there's no audit trail and no way to cut off access when someone leaves. A policy people can follow is worth more than one they ignore. This group closes the gap in 90 days — then dissolves.
The near-miss
This nearly started with us.
During the contract review itself — the work meant to catch exactly this — a Collage contract was put into a personal AI account, after a careful read concluded it held nothing confidential. It did. And the person doing it was the one paying the closest attention, on the single document they were specifically checking for this risk.
If it can happen there — on the one file under the most scrutiny — it's happening across the company. And no one is reporting it, because no one knows it's reportable.
That's not an argument for banning the tools. It's the argument for managed accounts, clear tiers, and a rule people can actually follow — which is exactly what this group is here to build.
The work
What this group delivers
Four deliverables. This is a group that ships, not one that meets.
A completed contract reviewEvery active project checked for the obligations that restrict AI, distilled into a short list people can actually check before a job.
The AI Use Policy v1.0The full policy plus a one-page quick reference, reviewed by counsel and HR. The document that governs all of it.
A rollout & training planHow the policy reaches every employee and every new hire — so it's followed, not filed.
The draft you'll pressure-test
How AI works here, in four lanes
The whole draft policy compresses to this. Permissive by design — with guardrails people can actually follow.
Green
Any reasonable tool.
- Public information; code & product research
- General questions
- Writing help with no project, client, or employee specifics
Yellow
Company-approved tools only.
- Internal SOPs, templates & checklists
- Internal notes & email
- Schedules & internal cost history
- Sub contacts & performance notes
Red
Approved tools only — and check the contract first.
- Drawings, specs & bid documents
- Bid & subcontractor pricing
- Contracts & change orders
- Client financial or donor information
- Employee personal information
- Security-sensitive airport, school & government documents
- Anything tied to a claim or dispute
Never
No exceptions.
- AI meeting bots — Otter, Fireflies, Read.ai
- Unverified AI numbers in a bid
- AI as your authority on code, law, or contract language
- Personal AI accounts for anything Yellow or Red
- AI behind a hiring, discipline, or safety decision
The whole policy in one line: project data goes in company tools; everything else, use judgment. And whoever signs it, sends it, or submits it owns it — AI doesn't change that.
Why you
Why you're in the room
Chosen by role and exposure — not by enthusiasm for AI.
A policy written only by people who love these tools will be reckless. One written only by skeptics will be useless. Good ones come from people who know the actual work — where the pressure is, where mistakes get expensive, where an hour saved matters. No policy written entirely by office staff survives contact with a jobsite.
Chair · PreconstructionOwns the policy and the deliverables — the single point of accountability.
Field operationsThe biggest visibility gap Collage has. What actually happens on a jobsite — not what the office assumes.
Preconstruction / EstimatingWhere the estimate and bid-review requirements land in daily practice.
Business developmentThe client-facing view — proposals, client disclosure, and how AI shows up in the work we put in front of owners.
Designated skepticSomeone who thinks this is overhyped or risky. Finds the holes before an owner or a bonding company does. Their sign-off is worth more than any advocate's.
That's your seat. Your job is to keep this grounded in how Collage really operates — and to catch the blind spots the rest of us will miss. If the skeptic never disagrees in a meeting, the wrong person is in the seat.
What to expect
What you're saying yes to
A real working group — not a standing committee.
- 01This intake — five minutes, now.It shapes the very first agenda, so the group starts on what you actually care about.
- 02A 90-minute kickoff.Charter, the contract-review findings, and a section-by-section challenge of the draft policy. The skeptic goes first. It's the only meeting longer than an hour.
- 03Six working sessions — biweekly, 60 minutes, hard stop.Agenda and materials circulated 48 hours ahead. No material, no meeting. The field seat can join remotely.
- 04Four deliverables — then it's done.The survey, the contract review, the policy v1.0, and the rollout plan. The group dissolves on delivery, or at 90 days.
Realistically: a handful of hours across about three months. In exchange, a policy that reflects how Collage actually works — because the people who do the work wrote it. Then the group stands down. It doesn't become a standing committee.
Before the first session
Tell us where you stand
Five minutes. This shapes the agenda around what the group actually knows and cares about — so we don't waste the first meeting getting oriented. When you finish, you'll get something back.
Analysis complete
Everything you just read was written for you, in your browser, from the answers you gave a moment ago — no server, no AI service, nothing sent anywhere. Which raises a fair question: who built all of this?
One more thing
How this page was made
This entire page — the writing, the design, the motion, and the read you just got back — was built by AI, working from a short brief, in a matter of hours. Not a template. Not a team of designers. A conversation.
That's not a gimmick. It's the exact capability this task force exists to point at the real work of Collage — carefully, on purpose, with the guardrails you're going to help set.