In an architecture firm AI arrives in five places before anyone decides anything: the design studio that generates renderings and summarises the client's programme, the production team that drafts the notes and checks the sheets against each other, the specifications group that reuses last year's sections, the proposal desk that writes the fee proposal and the qualifications package, and the contract administrators who draft the answer to the contractor's request for information and the field report. The principal's job is not to pick the tool. It is to decide three things before a tool touches a client's drawings, a rendering or a fee proposal: who in the firm owns the question, what may never go into an outside tool, and what the firm will never let a machine do alone. This article is for the principal, partner or owner of an architecture practice with a few dozen to a few hundred people, independent or newly backed by private equity, who is asking how to start with AI safely. The same three decisions for a consulting engineering firm, where the sealed drawing and the liability file shape the never-build list, are in AI for an engineering firm.
I was COO and then President of a public company that closed more than forty acquisitions, and I ran the integration on most of them. I am a lawyer by training. Now I write fixed-fee AI plans for CEOs of private-equity-backed and owner-led mid-market companies through 013 Labs, and architecture practices are a group where a tool's mistake can end up on a sealed drawing, in a client's building and in a liability claim at the same time.
Where AI already is in your firm
Ask a senior designer what the free tools on the staff's phones are doing this week and the honest answer is: turning a sketch into a dozen renderings for the client meeting, summarising the client's programme and the zoning documents, drafting the narrative for a design review. Ask the production team and it is the general notes and the cross-check of door numbers against the schedule. Ask the specifications group and it is last project's sections rewritten for this one. Ask the proposal desk and it is the scope of services and the fee proposal assembled from the last one. Ask the contract administrators and it is the minutes, the first draft of the answer to the contractor's question, the field report written from a phone's notes. The drawing-software and project-management vendors each switched on an assistant with their last release. None of it was approved. Some of it has already shown a client a rendering that looks like a commitment, and some has carried a client's drawings, a site survey or a building-information file onto a vendor's server under no agreement at all.
The three decisions, in the order a firm makes them
Who owns the question. Not the IT provider, the software vendor or a new hire. In most firms it is the director of operations, the quality or risk principal or a senior associate who also keeps the professional-liability file, as they are the people the staff already ask when something new shows up. Give that person a written scope of half a page: what they may approve alone, what comes to the principals, when the page is reviewed. The article on whether your IT company can handle AI draws the line.
The data rule. One sentence. For an architecture firm it usually reads: no client's drawings, design files, programme, site survey, building-information file or renderings, no consultant's work, no fee information and nothing about an employee go into a tool the firm has not approved, and no tool is approved to hold project information without written vendor terms that say who owns what the tool produces and what the vendor may do with what it is given. The principal writes it. The client's information is held under the client's contract, and the vendor's terms are the firm's to sign.
The never-build list. The three things the firm will never let a machine do alone. For an architecture firm the list writes itself: no drawing, specification or code review is sealed, signed or issued without the licensed architect who is responsible for it having personally reviewed the work; no design claim to a client, and no rendering that shows something the firm has not designed, is presented as approved, compliant or buildable without a licensed architect's name on it; no proposal, fee, scope of services or additional-services price is sent by a tool. Everything else is a question of fit. These three are not.
What fits at fifty people, and what does not yet
The first project is the work your people already do by hand, the same way, on every job: the client's programme and the zoning documents read into a summary for the designer to check against the source, the first draft of the general notes for the production architect to finish, the minutes drafted from the project architect's notes, the first draft of the answer to a contractor's request for information for the architect of record. Renderings for early studies are a fair use if every image is labelled an illustration. The article on picking your first AI project has the five tests. What does not fit yet is anything that decides or seals: the code review, the egress or accessibility conclusion, the sealed sheet, the fee. A tool can prepare any of those for a licensed architect. It should not finish them. Answering a request for proposal is a similar draft-and-check job, and can AI write RFP responses covers who must own every claim.
The licensing board, the liability insurer, the client and the buyer will ask
The AI question reaches an architecture firm from four directions that have nothing to do with technology. The state licensing board and the profession's own rules ask it in their words: an architect who seals work is responsible for it, and a tool's draft is not a defence. The professional-liability insurer's renewal form asks whether automated tools produce design work or opinions in the firm's name, because work a tool produced is work the firm produced. The client asks it in the contract with a clause about outside tools and who owns what the firm delivers. And a buyer, a larger firm or private equity, asks it in diligence, because architecture and design practices are bought into platforms; the article on what buyers ask about AI in due diligence has that version, and what to tell your lender about AI covers the bank. One dated page answers all of them.
The engineering side of the same project has its own version of the question. The article on AI for an engineering firm covers the consulting engineers who sit beside you on the drawings, and the one on AI for a construction company covers the contractor who builds from them.
Picture a regional architecture firm of ninety people with two offices and a new private-equity partner. A design studio has been running a client's programme and its own sketches through a free image tool on the designers' own logins. The principal finds out when the client's counsel asks, during a dispute over what was promised at a design presentation, which tool made the renderings that went into the deck and who confirmed that what they showed could be built and then finds that one image showed a stair the code does not allow, that no licensed architect had seen it before the meeting, and that the client's programme and survey went into the same tool under terms that let the vendor keep them. The fix takes two weeks and a call to the liability carrier. The director of operations is named as owner. The principal writes the data rule: no client file, no consultant file, no employee file into an outside tool, and no tool without written vendor terms that say who owns the output. The never-build list goes on the page, starting with no design claim or rendering presented as approved without a licensed name on it. The free tool is replaced the same month, the client gets corrected images in writing from the architect of record, the files are deleted from the vendor's account with the deletion confirmed in writing, and the client's counsel gets the page. This is a hypothesis, a composite and not a client.
The one page
What is in use, by studio and task, including the free tools and the assistants the drawing-software and project-management vendors switched on. Who owns the question, with the written scope. The data rule in one sentence, with the vendor terms listed and a line on who owns what each tool produces. The never-build list, the three items above. The review date and who signs. The principal signs it, and the director of operations has it ready for the insurer's renewal. The board-briefing template has the same page in a form a firm can fill in.
Mistakes that repeat
Letting the software vendor switch it on. An assistant arrives in a release, defaults to on, and the next set of general notes carries a statement nobody checked against the design. The principal decides, feature by feature.
The client's files in a free tool. The mistake that puts the firm in front of the client's counsel and, if work is sealed on the strength of it, the licensing board. The data rule exists for this.
A rendering that reads as a promise. A tool produces a handsome image of a building that was never designed to that standard, the client falls for it, and the fee, the code and the budget have to catch up. Every image is labelled as an illustration, and a licensed architect signs off on anything a client could take as a commitment.
Terms nobody read. The vendor's page lets it keep what it is given, or is silent on who owns what it produces, and the firm finds out when a client asks for its drawings back.
A proposal the template wrote. A tool rolls last year's fee proposal forward with the dates changed; the scope has changed, the fee does not cover it and nobody noticed. A principal reads every proposal before it goes.
No review date. Clients and insurers change their language. A page without a date is a page nobody opens.
Frequently asked questions
Should an architecture firm use AI?
It already does. The decision worth making is who owns the question, what may never go into an outside tool, and the three things the firm never lets a machine do alone: a drawing or code review sealed unread, a design claim or rendering presented as approved without a licensed architect, and a proposal or fee sent by a tool.
How can an architecture firm start with AI safely?
Start with the page, not the tool. Name the person who owns the question, write the data rule with client files first and written vendor terms as a condition, and put the never-build list on paper. Then pick one repeatable job, such as meeting minutes, and run it inside the rule for a month.
Who should own AI in an architecture practice?
The person the staff already ask when a new tool shows up, usually the director of operations, the quality or risk principal or the senior associate who holds the liability file, with a written scope of half a page and a review date. Not the IT provider and not the drawing-software vendor.
Can AI generate renderings and draft specifications for us?
It can produce early-study images and a first draft of a specification section from documents a person has given it, for the designer or the specifier to check against the design, the code and the project. It should not make the call. Every image is labelled as an illustration, and the seal is the architect's; the client's counsel or the board will ask who checked it.
What about client drawings and design files in these tools?
They are first in the data rule. No client's drawings, design files, programme or survey go into a tool the firm has not approved, and no tool is approved to hold project information without written vendor terms that say who owns what the tool produces. If client files have already gone into a free tool, delete them from the vendor's account, get the deletion confirmed in writing, and tell the firm's liability carrier the same week.
Why does my insurer or my client ask about AI?
Because each carries the risk when a drawing, a rendering or an answer that a tool touched turns out to be wrong, and each wants to know whether a licensed architect stands behind it. One dated page with names on it is the answer, and the same page answers the licensing board, a buyer and the bank.
Not sure where you stand? Take the 3-minute AI readiness scorecard.