AI Training for Architects: What a Practice Workshop Covers and Costs

A practice-wide AI workshop for a small architecture practice should cover the practice's own rules for client data, writing and admin on real documents, research checked at source, and image tools. For twelve people, a half-day costs about $400 to $700 a person, assuming a $1,500 to $5,000 session fee and staff time at $60 an hour.

Those fees are my assumptions, not quotes: none of the providers I checked prints a price for a practice session, so replace them with the quotes you collect. Search results also answer a different question. They are mostly individual courses: a six-week live course on design workflow, short recorded courses, and a three-day programme titled "AI for Architects" that is written for software architects. Provider details below come from each provider's own page, read on 29 October 2026, and where a page prints no price I say so.

Follow me on Instagram@sagnikteaches

What the first page of results is selling, and to whom

An individual course teaches one person a skill. A practice workshop has a different job: agree the rules, practise on the practice's own documents and leave every person with two or three tasks they will actually do. The table sorts the routes you will meet.

Connect on LinkedInSagnik Bhattacharya

Two lessons come out of the table. Check the audience line before you pay: the three-day course with the right title is for people who build AI systems, which is why it is the wrong answer to this question. And notice how a team workshop is priced even when the amount is hidden: 13Labs calls its four-hour Kickstart a "fixed price for up to 10 team members", per session and for a capped group, not per person. So the figure you need to ask a provider for is a session fee and a group cap.

Subscribe on YouTube@sagnikteaches

What a practice-wide half-day should cover, hour by hour

This is the agenda I would ask a provider to confirm in writing. It runs four hours, uses the practice's own anonymised documents and ends with each person owning two tasks. The structure borrows from the agendas that providers publish, such as 13Labs' four hours covering AI in your industry, prompting, building first automations and integration, and Gendo's list of fundamentals, privacy and governance, visualisation workflows and authorship. The difference is that the practice's own rules come first.

Notice what is missing. There is no tour of every tool and no lecture on how models work. Each block ends in something the practice keeps. The longest one is the writing block, because writing and admin are where a small practice finds its first dependable hours, as six jobs where AI helps a small practice sets out.

Modules by role: who needs which extra hour

The half-day is common ground. After it, each role gets a short module on its own work. This keeps the half-day affordable and stops a principal sitting through BIM scripting.

The ELVTR syllabus is a useful contrast, because it covers what an individual designer would want: concept generation and storytelling with image tools, ethics and regulation, AI in BIM workflows and digital twins. That is valuable for one person's craft. It does not cover what a practice needs to settle together, such as whose accounts are used and what leaves the building.

What a half-day costs per person for a twelve-person practice

The main worked example uses a practice of twelve people: two principals, five architects, two technologists, a visualiser, a practice manager and an administrator. Staff time is valued at a blended $60 an hour, a placeholder you should replace. Everyone spends 4 hours in the workshop and half an hour on pre-work, which is 4.5 hours, or $3,240 for twelve people. Four 30-minute follow-up check-ins add 2 hours each, $1,440 in all.

The provider's fee is the number I cannot supply, because none of the pages I read printed one for a practice-wide session. Treat the three fees in the table as placeholders for quotes you collect, not as market rates.

That produces the $400 to $700 in the opening answer, which is the range without check-ins. Staff time is $270 a head in every column (more once check-ins are added), and it is larger than the fee share until the fee passes $3,240. Group size changes the fee share fast: the same $1,500 session for six people is $250 a head. So ask what the cap is, and fill the room. A six-person studio should compare quotes on per-session price and on whether the provider will mix it with another small practice.

Self-taught or workshop: the time arithmetic

Self-taught is not free. Suppose each person takes one recorded course of about 3.5 hours and spends 4 hours practising on real work, which is 7.5 hours. At $60 an hour that is $450 a person and $5,400 for twelve, before any course fee, and for that you get twelve different habits and no shared rules. A free vendor session of two hours costs $120 a head in time, or $240 with four check-ins, and the practice still has to write its own rules.

The workshop row's $390 is the 4.5 hours plus the four check-ins: $270 + $120. Cash apart, the routes land close together: self-taught is $450 of time plus the course price, and the workshop is $515 to $807 all-in. The difference is what you get. The workshop buys shared rules and one set of habits for a practice of ten or more; self-taught buys flexibility. For a two-person studio the session fee spread across two people is steep, and a self-taught route with a shared weekly hour is enough. How much time and money AI staff training takes gives wider ranges across business types.

The rules page a workshop should leave you with

If the half-day produces one thing, make it this page. It is a draft for partners to edit, not legal advice, and any contract confidentiality clause or insurer condition should be checked against it. An AI policy checklist and clauses for architects goes deeper.

Practice AI use, one page (draft for partner sign-off)
1 Accounts: practice business accounts only; no personal accounts for client work.
2 Never enter: unreleased client schemes, competition entries, drawings under
  NDA, personal data about clients or contractors, fee rates for named clients.
3 May enter, with care: anonymised meeting notes, our own boilerplate, public documents.
4 Always check: every standard, clause number, dimension and legal reference
  against the primary source before it leaves the office.
5 Images: concept use only, labelled "AI-assisted concept" in client material;
  no AI image shown as design intent without a drawing behind it.
6 Record: note AI use on the project file when AI text or imagery goes to a client.
7 Owner: practice manager. Review every six months or when a tool changes its terms.

Line 2 is the one that matters most when a tool is free or personal. Many consumer plans let people switch off use of their chats for model training, but they sit on personal accounts the practice cannot see or control. Business plans for ChatGPT and Claude both state that content is not used for model training by default, which is one reason line 1 exists. If you have customers in the EU, the AI Act also asks organisations that use AI to take measures supporting staff literacy; a documented half-day with check-ins is one such measure, and a qualified adviser can say what more your situation needs.

A drafting exercise: a prompt, a result and what to fix

The writing block works best with a task every person recognises. This is a fee-proposal paragraph for an invented small extension, and the prompt keeps the AI inside the facts.

You help a small architecture practice write fee proposals. Use only the
facts below. Write the scope paragraph for the concept design stage in plain
English, about 120 words, then a separate list of exclusions. Do not mention
fees, dates or standards.

Facts: two-storey rear extension to an older house. Client wants an open-plan
kitchen and a home office. Included: two design options, one client meeting
per option, sketch plans and a 3D massing view. Excluded: structural design,
drawings for the consent application, interior design, neighbour agreements.

A typical result (illustrative): "At concept stage we will prepare two design options for the proposed two-storey rear extension, each shown as sketch plans and a simple 3D massing view, and discuss each with you in a meeting. Exclusions: structural design, drawings for the consent application, interior design and neighbour agreements." What you would fix: models often add inclusions the facts never gave, such as "up to three rounds of revisions" or "a feasibility study", and sometimes soften an exclusion. Compare the output line by line with your terms. Mark each added phrase, and decide in the room whether the workshop's rule is "delete it" or "ask the partner". The drafting habit this builds, writing from facts and checking against terms, is the one that transfers to briefs, specifications and minutes. AI for fee proposals, briefs and paperwork extends it.

Research you can trust: a test with a known answer

The research block needs a test the team cannot fail quietly. Ask each person to put a question to an assistant where they already know the answer from their own work, such as a minimum clear width for an escape route or a dimension for a guarding height. Then they open the primary source. The point is not to catch the AI out. It is seeing how confident a wrong answer sounds. A common slip is a plausible number that belongs to another set of rules or an old edition, delivered without a clause reference. Record every wrong or unsourced answer on a flipchart. The rule that follows, written into line 4 of the one-page rules, is that no figure leaves the office until someone has read it at source.

Tool seats after the workshop: a monthly line, not a surprise

The workshop's cost is one-off. Seats are monthly. Decide them after people have tried the tools on two real tasks, then buy for those who used them.

One year of a Standard seat is $240, which is less than half the $515 a head the half-day costs with check-ins. That ratio is a useful sanity check. If training costs several times the annual price of the seat, the training had better change how the practice works. For help choosing between those assistants, Claude, ChatGPT or Gemini for architects compares them on practice tasks.

The four-week follow-up: two tasks each, logged

The half-day only pays if people keep using what they practised, so the follow-up is part of the cost and part of the proof. Each person logs their two tasks for four weeks: minutes before, minutes after including the time spent checking, and how often the task comes up. A task that is not faster after checking is dropped, which is the honest outcome. This is an invented tracker for six of the twelve people.

The six rows add up to 940 minutes, about 15.7 hours, or $940 a month at $60 an hour. Take off $160 for eight tool seats and the net is $780 a month. Divide each total cost by that to get the payback: the $1,500 session at $6,180 pays back in 7.9 months, the $3,000 one at $7,680 in 9.8 months, and the $5,000 one at $9,680 in 12.4 months. These are made-up logs, so the lesson is the method. If your real numbers show a payback beyond a year, shrink the next step, because a smaller group on a free session, or a one-hour team session you run yourself, may be enough to start.

What goes wrong when the workshop skips the rules

A practice books a workshop built around image generation because the visualiser is excited. The session is a hit. Three weeks later a junior architect, wanting a quick massing variation for a competition entry, uploads the unreleased scheme to a personal free account. Nothing visible happens. The problem surfaces months later, when the practice's insurer asks how AI is controlled and the answer is that nobody wrote anything down.

It shows up as a question with no answer, not as a failure. The workshop was good at what it covered and never covered the rule. The fix costs less than the workshop: run the first 30 minutes of the agenda above, get the one-page rules signed, switch the team to business accounts and add a line to the project file. It is worth telling your insurer early, before they ask.

Judging a quote before you book

When quotes arrive, compare them on the same eight questions. Here is a filled-in comparison of three invented providers for the twelve-person practice.

The cheapest row is not the winner. B misses two things the practice needs, its own documents and a rules page, so a cheaper fee buys a worse day. A is the better value even though its price is higher. Check how each figure is built before you trust it: A's $470 is $200 of fee share plus $270 of staff time, B's $360 is $150 plus $210 for its shorter day, and C's $562 is $292 plus $270, because C's fee covers fifteen seats and only twelve people attend. For the wider choice of providers, how to choose an AI training provider for your team adds questions about credentials and references, and a first-quarter pilot plan, linked at the end, shows where the half-day sits in a longer plan.

Questions practices ask before booking AI training

Is a free vendor session worth taking?

Usually, for the agenda, if you treat it as a sales-adjacent session. Gendo's page offers a free one-to-two-hour CPD session for studios covering fundamentals, privacy and governance, visualisation workflows and authorship, led by its chief executive. Expect it to centre on its own product, and still write your own policy afterwards.

Does the AI Act affect training for a small practice?

Only if you have customers in the EU. Its AI literacy duty asks organisations using AI to take measures supporting staff literacy, and a 2026 amendment softened the wording so that no specific level has to be guaranteed. A recorded half-day and a follow-up show measures taken. Ask a qualified adviser what your own situation needs.

Should the whole practice attend, or just the partners?

Everyone who will touch client documents or images should attend the common half-day, because the practice rules apply to all of them. Add short role modules afterwards. Partners need an extra hour on contracts, insurer questions and tool budgets, since they decide what is bought and what is disclosed to clients.

How soon should the practice buy tool seats?

After the workshop, not before. Let each person try the tools on two real tasks first, then buy seats for those who used them. Business seats for ChatGPT and Claude both list at $20 a month on annual billing or $25 monthly, with a two-seat minimum, so eight seats cost about $160 to $200 a month.

Further reads

• Copyright and Client Data Risks of AI in Architecture Practices — The copyright and client-data risks a workshop should cover.
• Best AI Rendering Tools for Small Architecture Practices (2026) — Rendering tools priced for a small practice.
• How to Run a 60-Minute AI Workshop for Your Team — A one-hour version you can run yourself.
• How Much Does AI Cost a Small Architecture Practice per Month? — What AI costs a small practice each month.
• AI Consultant for Architects: What a First Engagement Costs — What a first outside engagement costs.
• AI Training for Construction Companies: Workshop or Self-Taught? — Work out AI training cost per head by role for a building firm: workshop fee, paid hours off site, licences and a 90-minute pilot.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ELVTR AI for Architects course page; a university training centre's AI for Architects course page; Gendo free CPD page; 13Labs AI Kickstart page; MyArchitectAI pricing page; OpenAI ChatGPT Business pricing page; Claude pricing page; Microsoft 365 Copilot Business page. All read on 29 October 2026. Practice size, hourly cost and provider fees in the examples are placeholders, not market data.