What AI Is Good for Architects? Six Jobs for a Small Practice

A small architecture practice gets most from three tools: a general assistant on a business plan, a meeting note-taker and one concept-image tool. Use them for six jobs: client briefs, planning statements, meeting notes, fee proposals, consultant coordination and concept visuals. Start with meeting notes, then briefs, and keep snagging lists away from AI.

The common thread is text and early-stage images. AI is strong where the work is reading, restructuring and drafting words from material you supply, and weaker where the output has to match a physical building you can see and nobody else can. Each job below has a workflow, a worked example and a note on where it goes wrong.

Follow me on Instagram@sagnikteaches

1. Turning a client conversation into a written brief

A first meeting with a client produces a rush of wishes, budgets and half-decisions. Most practices write it up afterwards from memory and handwritten notes, and then discover in month three that the client remembers it differently. A written brief fixes that, and an assistant can produce a clean first version from your raw notes or a transcript.

Connect on LinkedInSagnik Bhattacharya

The workflow is short. Record the meeting only with the client's agreement, or type your notes straight after. Paste the notes into a business-plan assistant with an instruction that separates what the client said from what you inferred.

Subscribe on YouTube@sagnikteaches

Turn these notes into a one-page client brief for an architect.
Headings: Project and site; Must-haves; Nice-to-haves; Budget stated; Programme stated; Constraints mentioned; Open questions.
Only include what is in the notes. Mark anything I inferred with [inferred]. List what is missing as questions to ask the client.

Say a four-person practice meets a couple about a rear extension. The notes mention a kitchen-diner, a budget of "around 120,000", a wish to finish before the following winter, and a neighbour who objected to a previous scheme. An illustrative output lists the must-haves, records the budget as a stated figure without saying whether it includes sales tax or fees (the notes did not say), and asks whether the budget includes fees, furniture and landscaping. That last question is the value: the tool surfaces the gaps while the client is still on the phone to answer them. What you would fix is any invented detail, such as a stated completion month nobody gave. The [inferred] tag exists to make that visible.

Send the brief back to the client for confirmation as a short email. That single step avoids more disputes than any template. If you want the same idea applied further down the chain, how architects use AI for fee proposals, briefs and paperwork covers the wider paperwork set.

2. Drafting planning and permit statements from your own notes

Design statements, heritage statements and planning cover letters are structured, repetitive and written to a known set of headings. They are also the documents where a generic AI paragraph reads as padding, so the job is to draft from your design reasoning, not from nothing.

Feed the assistant three things: your design rationale in bullet form, the relevant local policy extracts you have already read, and an outline of the headings your authority expects. Ask it to write each section only from those inputs.

Draft the "Scale and massing" section of a planning statement (about 250 words, plain English, third person).
Use only these points: the extension is single storey, 3.2m to eaves, set back 1.5m from the boundary; the rear garden remains 11m deep; the roof matches the existing pitch; materials are brick to match with zinc cladding to the flat section.
Refer to the policy extracts below by their reference only. Do not add policy references that are not in the extracts.
[paste extracts]

An illustrative output will read smoothly. What you fix: invented policy numbers (the tool will happily cite a plausible one), claims such as "will not impact neighbouring amenity" that you have not assessed, and adjectives that make a modest scheme sound grand. Every policy reference and every dimension is checked against your drawings and the actual policy documents before the statement goes in. A realistic mistake: a draft said the extension was "subordinate to the host dwelling" when the proposed ridge matched the existing one. The architect caught it on read-through.

Because the rules and expectations differ by authority, keep your own template of the headings and let the assistant fill them in. The saving is typically the first draft of 1,500 words, which might take half a day, produced in a few minutes and then edited for an hour. Do not use it to invent justification for a scheme that does not meet policy.

3. Meeting notes and action lists

This is the easiest first win, and the one with the clearest payback. A design team meeting with a client, structural engineer and contractor yields decisions that nobody writes down well. An AI note-taker or a recording plus transcript tool produces a record, and an assistant turns it into actions.

Tools differ on price and limits. Otter's Pro plan is listed at $8.33 a user a month billed annually ($16.99 monthly), with 1,200 transcription minutes a user a month and a 90-minute cap on each conversation, so a long design review may need splitting or the Business plan's longer limit. Fireflies and the transcription built into video calls are alternatives. Always tell attendees you are recording and get agreement, since some clients and consultants will refuse, and recording rules vary by place.

From this transcript, produce: (1) decisions made, each with who decided; (2) actions with owner and date; (3) items deferred to the next meeting; (4) risks anyone raised.
Use a table for actions. If an owner or date was not stated, write "unassigned" or "no date given", do not guess.

A filled-in example: after a site meeting, the output lists "Structural engineer to confirm lintel depth over the new opening, no date given" and "Client to choose flooring by Friday, owner: client". Two checks make this safe. First, read the decisions list against your memory before sending, because transcription can swap "do" and "don't". Second, circulate it as "draft minutes for comment" so the others can correct it. A practice that sends notes within an hour of a meeting, rather than three days later, usually gets its actions done sooner, which is worth more than the minutes saved writing.

4. Fee proposals that match the scope

A fee proposal is a scoping document disguised as a letter. AI is useful because the weak point is consistency: scope items, exclusions, stage definitions and payment terms need to match from one proposal to the next. Give the assistant your last three good proposals (client names removed) as precedent, plus the new brief from job 1.

Using the two example proposals as precedent for structure and tone, draft a fee proposal for the brief below.
Include: scope by stage, exclusions, assumptions, client responsibilities, payment stages as percentages, and a list of what would be an additional service.
Use [fee] placeholders for all money figures. Do not invent fees.

The placeholders matter. Fees are your judgement of risk and time, and the model has no basis for them. A worked sum, illustrative: a rear extension with a stated budget of 120,000 and a fee of 9 percent implies a fee of 10,800. You set the percentage, then ask the assistant to split it across stages as 15, 25, 30, 20 and 10 percent and show the amounts. It handles that arithmetic cleanly, but you check it, because assistants occasionally round differently and the stages must total exactly.

The most valuable output is the exclusions list. A frequent cause of unpaid extra work is a scope that never said "party wall surveyor fees are excluded" or "interior design beyond layout is excluded". Ask the assistant: "list ten things a client commonly assumes are included in this stage that are not, based on the scope above." Choose which belong in the proposal. Once the proposal is accepted, the same scope and stage percentages feed the fee invoices, so keep them in one place.

5. Consultant coordination logs and chasing

A project has an engineer, a surveyor, perhaps a heritage consultant, a quantity surveyor and the contractor. The coordination job is keeping one list of who owes what, and chasing politely. AI helps in two ways: building the log from emails and minutes, and drafting chasers in the practice's voice.

Say you paste a week of project emails into the assistant and ask for a table: item, who is waiting, who owes it, date first asked, days outstanding. The first run is rough, but it catches things nobody had noticed, such as a drainage survey requested three weeks ago and never answered. Then ask for a chaser for each outstanding item:

Write a short, friendly chaser to [first name] at the structural engineer's office about the outstanding lintel calculation first requested on [date]. Say we need it by [date] to issue the building control or permit package. Two sentences, no pressure, no blame.

A before and after: before, the project architect kept the log in her head and a half-updated spreadsheet; after, a weekly paste-and-prompt takes ten minutes and the log is current. The limit is confidentiality. Do not paste client personal data or commercially sensitive tender numbers into a general tool; strip them first or use your business plan. And do not let it send anything: it drafts, you press send. If you want to automate the reminders themselves, that is a case for your project software's reminders rather than a chatbot.

6. Concept visuals and moodboards

This is the job most people think of first and the one where expectations most often outrun results. Image tools can produce atmosphere quickly: a sketch-to-render pass over a massing model, a moodboard of materials, a quick look at three facade tones. Veras, an architecture-focused tool now part of Chaos, works on top of your 3D model; on annual billing its Starter licence is listed at $202.80 a year and Pro at $348 a year, with monthly billing costing more and a 14-day free trial. Midjourney, a general image tool, runs from $10 to $120 a month across four plans, and only its Pro and Mega plans include Stealth Mode, which keeps your images out of public view, a point that matters for unpublished client schemes. Prices and trial terms change, so confirm them on the vendor's page. For a broader comparison see best AI rendering tools for small practices.

The workflow with the least risk is to treat visuals as conversation starters. Export a clean massing view, apply a style prompt such as "brick extension with zinc roof, overcast daylight, residential garden", and generate four options. Show the client as "concept atmosphere, not a design proposal". A before and after: a practice that previously commissioned an outside visualiser at several hundred dollars per image for a first concept now generates three variants in an afternoon, then commissions a proper render once a scheme is chosen. The comparison between doing both is covered in AI renders versus outsourced visualisation.

Where it goes wrong: generated images add windows, change proportions and invent details, so the picture can promise something the design does not contain. Label every image, keep the massing model as the source of truth, and never send a generated image as a planning document. Copyright and ownership of AI output are unsettled, which is why copyright and client data risks for practices is worth reading before images go to a client or a planning officer.

The one job to keep away from AI: snagging lists

A snagging or defects list records what is physically wrong on site, room by room, with photographs and a reference to the specification. It is tempting, because it is a list of items and photographs, but it is the wrong job for a general tool for three reasons.

First, the value is in what you saw. An assistant cannot judge whether a paint finish meets the specification from a phone photo taken in poor light, and it will happily describe a defect that is not there or miss one that is. Second, the list may become a contractual record in a dispute, so every line needs a named person who looked at that wall. Third, the cost of a missed item lands on the client after practical completion.

A permitted use is narrow and clerical: once you have written the defects in your own words, an assistant can tidy them into consistent wording, number them and sort them by room or trade. The inspection itself, and the judgement of severity, stay with you. A realistic mistake: a practice asked a tool to "list likely defects in a new-build bathroom" and then used the output as a checklist. The tool listed generic items and the inspector, anchored by the list, missed a poorly bedded shower tray.

Choosing which job to start with

Start with meeting notes (job 3), because it fails cheaply and you will see the benefit within a week. Add the written brief (job 1) next, since it uses the same skills. Fee proposals (4) and statements (2) come after you trust your checking habits. Visuals (6) can run in parallel if someone in the practice enjoys it, but keep it out of client deliverables for now.

A rough monthly cost for a small practice: a business-plan assistant at $20 to $25 a seat a month (annual or monthly billing, two-seat minimum), one note-taker plan at under $20, and optionally one image tool at roughly $17 to $60. That is roughly $70 to $150 a month for a three-person team, before the time to learn. The time cost matters more, so give one person two hours a week for the first month and ask them to record minutes saved on each task. How much AI costs a small architecture practice per month breaks the cost down by tool.

Which assistant to buy is a smaller question than which job to give it. If you want to compare, Claude, ChatGPT and Gemini for architects sets them side by side. If the comparison leaves you between two, run the same meeting transcript and the same brief through both for a week and keep the one whose drafts need fewer corrections.

A first-week plan for the busiest principal

Principals rarely have a free afternoon, so here is a plan that fits around live projects. On Monday, choose one live job with a meeting that week and agree with the client that you will take notes with a tool. On Tuesday, set up one business-plan assistant, switch off training on your content where the plan allows, and write the one-page prompt for meeting notes in a shared document so everyone uses the same wording. On Wednesday, run the meeting, paste the transcript, and read the output against your own handwritten notes. On Thursday, send draft minutes for comment and count how many corrections come back. On Friday, spend fifteen minutes writing down what the tool got wrong and what you changed in the prompt.

By the second week you will know whether the tool suits the way you talk and the way your consultants speak. Strong accents, background noise on site and overlapping speech all lower transcript quality, so test on a quiet office meeting before a noisy site visit. A realistic mistake: a practice trialled the note-taker on a site meeting beside a cement mixer, got a transcript full of gaps, and wrongly concluded the whole approach did not work. The fix was to record the office review meeting afterwards and dictate the site actions separately.

Use the same first-week approach for a second job. If notes went well, take the brief next, using last week's meeting as the raw material. Because the brief and the minutes come from the same conversation, you can compare the two and see whether anything was recorded in one and missed in the other. That cross-check is a cheap way to find the tool's blind spots before it touches a fee proposal.

Client expectations and what to say about AI use

Clients increasingly ask whether a practice uses AI, sometimes because they are curious and sometimes because they are worried about confidentiality or about being charged for work a machine did. A short, honest line in your appointment terms or fee proposal handles most of it: that you may use AI tools for drafting, note-taking and early concept images, that a named architect reviews everything issued, and that client information is not used to train public models. Check each part of that sentence against the plan you actually use before you promise it.

On fees, the question is whether time saved should reduce the price. Many practices fix the fee by stage and scope, which makes the question mostly moot, and keep the efficiency as margin. If you charge by the hour, be transparent: bill the time you spent, including checking, not the time the work would have taken without the tool.

How to check each job is working

Keep it light. For each job, record three things in a shared note: minutes before, minutes after, and the number of corrections you made. After a month, anything that saved time with few corrections stays; anything with many corrections goes back to the prompt or off the list. A fee proposal that needs a full rewrite each time is telling you the precedent documents are inconsistent, not that AI is useless.

A short worked tally, illustrative: a small practice tracks 12 meeting notes, 3 briefs, 2 fee proposals and 1 statement in a month. Notes drop from 35 minutes to 10 including review, a saving of 25 minutes each, or 5 hours. Briefs save 40 minutes each. Fee proposals save about an hour each because exclusions are drafted for them. The statement saves two hours. That is around 10 hours across the month, against roughly $90 in tool costs and a couple of hours of set-up. Your figures will differ, which is the reason to measure them.

If you would rather map this with someone, that is what my AI implementation consultation covers: finding the jobs where AI pays off in your practice and handing over something the team can run.

Questions from practice principals

Do I need an architecture-specific AI tool?

Not for five of the six jobs. A general assistant on a business plan covers briefs, statements, notes, fee proposals and coordination logs. Architecture-specific tools matter mainly for concept visuals, where image quality and control are the point.

Is it safe to upload drawings to a chatbot?

Treat drawings and client documents as confidential. Use a business plan whose terms say your content is not used for training, upload only what the task needs, and check your appointment terms and any consultant confidentiality clauses first.

Who owns AI-generated concept images?

Ownership and copyright for AI output are unsettled and vary by place and by tool terms. Use generated images for early concepts and moodboards, label them, and do not present them as a finished design or as the practice's authored work without advice.

How much time can a small practice realistically save?

Expect modest gains at first. Meeting notes and first drafts of text usually save the most, while visuals save little unless you already know the style you want. The tally in this tutorial is illustrative only, so time five tasks before and after to find your own figure.

Further reads

• AI Floor Plan Generation for Architects: Where It Fails a Real Brief — Why generated floor plans fail real briefs.
• Is ChatGPT Plus Worth It for an Architect? A Practice Cost Check — A cost check on one paid assistant plan.
• Best AI for Architecture and Interior Design: Concept Image Tools by Cost — Concept image tools compared by cost.
• AI Consultant for Architects: What a First Engagement Costs — What a first AI engagement for a practice costs.
• AI Construction Design Software: What Renovation Contractors Can Draw — What AI design software can draw for builders.
• AI Automation for Architects: Enquiry to Fee Invoice, Step by Step — Connect enquiry, proposal and fee invoice in one flow.
• Can AI Help a Small Architecture Practice? Where It Saves Time — The wider case for AI in a small practice.
• AI Training for Architects: What a Practice Workshop Covers and Costs — A half-day AI workshop agenda for an architecture practice, modules by role, and the cost per person once staff time is counted, set against self-taught routes.
• AI Marketing for Architects: Case Studies, Website Copy and LinkedIn — Turn a site-visit voice note into a case study, service-page copy and LinkedIn posts that sound like your practice, with permission and claims checks.
• Architect Meeting Minutes: Record, Summarise and Issue Same Day With AI — A same-day workflow for architects: consent to record, a minutes template, a decisions log and an issue sheet, using Otter, Teams Copilot or Meet notes.
• Architect Website: Project Descriptions and Service Pages Written With AI — A copy workflow for architects: project descriptions from verified facts, service pages by project type, honest area pages and an FAQ on fees and process.
• How Can AI Help Architects? Briefs, Planning Notes, Consultant Chasers — Admin jobs beyond rendering: brief summaries, planning-note drafts and consultant chasers, sorted by how closely the architect must review each.
• How Can Architects Use AI on Planning Applications? Where to Draw the Line — Where AI helps on a planning or permit application, where it stops, and the checks an architect runs before anything is submitted.
• How to Use AI in an Architecture Firm: A First-Quarter Pilot Plan — Run three 13-week pilots for meeting minutes, fee proposals and site reports, with baselines, named KPIs, a weekly quality sample and a written go or stop rule.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI business pricing, Claude pricing, Otter pricing, the Veras page on chaos.com and Midjourney's plan comparison, all checked November 2026. Check current plans before buying.