AI for architects: fee proposals, specs, and planning admin in 2026

You spent four years at architecture school, two more in Part 2, and another year preparing for Part 3. You passed your ARB registration. You know how to design a building that stands up, looks good, and satisfies planning. Yet you spend Tuesday afternoon copying paragraphs from last year's fee proposal into this year's, adjusting the project name and square meterage, and hoping the client does not notice you recycled the same boilerplate about your 'collaborative approach to placemaking'.

Most architectural practices run on a combination of Revit or ArchiCAD for technical work, a jumble of spreadsheets for fee tracking, Word templates older than some of your Part 1 students, and a planning portal that feels like it was designed to punish optimism. The creative work is maybe 30 per cent of your week. The rest is fee proposals, pre-app submissions, NBS specification paragraphs, contractor queries, and client emails asking why the planning officer has gone quiet.

AI for architects in 2026 is not about generating elevations or automating design. It is about reclaiming the hours you lose to admin, proposal drafting, and repetitive document production so you can spend more time on the work you trained for. This guide covers how architecture practices are using AI for fee proposal drafting, NBS specification writing, planning portal admin, client communications, and integration with Revit and ArchiCAD, with attention to ARB and RIBA requirements.

What AI for architects actually means in practice

Most of the AI tooling marketed to architects right now falls into two categories: rendering engines that turn sketches into photorealistic images, and generative design plugins that optimise floor plans for daylighting or circulation. Both are interesting. Neither solves the problem that ate your Monday morning, which was writing a pre-application submission for a residential extension in a conservation area.

**Architecture practice AI that works is task-specific.** It drafts fee proposals based on your past projects and the RIBA Plan of Work stages you are quoting for. It writes NBS specification clauses by pulling from your office library and cross-referencing the materials you have specified in Revit. It monitors planning portal status updates and drafts holding emails to clients when the case officer requests another heritage statement. It does not try to design the building. It does the work that stops you designing the building.

The practices getting value from AI in 2026 are not the ones installing experimental plugins. They are the ones who identified the three or four repetitive admin tasks that consume ten hours a week, then built or commissioned agents to handle them. The return is not a better facade. The return is getting home before your kids go to bed, or being able to take on another project without hiring another associate.

Fee proposals: the task every practice hates and every client judges you on

A residential client emails on Thursday. They want a fee proposal by Monday for a two-storey rear extension, RIBA Stages 2 to 4, estimated construction value around £180,000, site in a conservation area. You have done a dozen similar projects. You know roughly what you will charge. You also know you will spend four hours over the weekend pulling together a proposal document because the last good template you had is three laptops ago.

**AI for architectural practices handles fee proposal drafting by treating your past projects as a corpus.** An agent trained on your won and lost proposals can generate a first draft in about eight minutes. It pulls scope descriptions from previous residential extensions, adjusts RIBA stage deliverables to match what you quoted last time for conservation area work, and inserts your standard terms lifted from the last CIC agreement you used. It does not invent a fee. It drafts everything except the number, which you set based on the conversation you had with the client.

The agent does not need to be sophisticated. It is a document assembly task with some light context retrieval. You feed it the client brief, point it at a folder of past proposals and your office CVs, and it returns a Word document with your branding and the right RIBA stage descriptions. You spend 30 minutes editing the scope to match what the client actually asked for, insert your fee, and send it. Four hours becomes 30 minutes. Over a year, that is 80 hours saved if you price ten projects a month.

Some practices connect this agent to their CRM so it can pull client history automatically. If you quoted the same client two years ago for a loft conversion, the agent can reference that project in the new proposal without you remembering it existed. That continuity makes you look more attentive than you have time to actually be, which is the point.

NBS specifications: the document nobody reads until the contractor queries it

You have specified a green roof. The contractor emails: what substrate depth, what fall, what membrane, what edge detail, and can they substitute the sedum mat supplier because the one you listed has a 12-week lead time? You open the NBS specification document you last touched in Stage 4, search for the green roof clause, and realise it says 'specialist green roof system, exact specification to be confirmed', because you were going to come back to it and then you did not.

**NBS specification writing is a retrieval and assembly problem.** You have specified green roofs before. You have a clause somewhere in an old project file. You know the substrate depth is 80mm for sedum and 150mm for wildflower mix. You know you prefer a particular fall because anything flatter ponds. You have this knowledge. You do not have it indexed in a way that lets you pull it into a new spec document in five minutes without opening four old Revit models to check what you did last time.

An AI agent for NBS specification drafting sits between your Revit model and your NBS library. When you specify a material or assembly in Revit, the agent searches your past projects for similar specifications, pulls the relevant clauses, and drafts an NBS paragraph that matches your office standards. If you have specified Kingspan K5 in six past projects, the agent knows your preferred thickness range, your thermal performance targets, and the edge detail you usually draw. It writes that into the spec. You review it, adjust for this project's specifics, and move on.

The agent can also flag inconsistencies. If your Revit model shows 100mm Kingspan but your NBS spec says 120mm, it tells you. If you have specified a product that you substituted out in the last project because of lead times, it surfaces that note. It does not make design decisions. It stops you from issuing a specification document that contradicts your drawings or repeats a mistake you already learned from.

Practices that have integrated this workflow report that their contractor query volume drops by about 40 per cent in the first six months, mostly because the specs are now complete and consistent enough that contractors do not need to ask. You still get queries. They are just about site conditions and buildability, not because you forgot to specify the roof membrane.

Planning portal admin: the digital black hole where optimism goes to die

You submitted a householder application three weeks ago. The planning portal says 'awaiting validation'. You email the case officer. No response. You call the planning department. The receptionist says the case officer is on leave. You email again. The case officer replies a week later: the red line boundary on your location plan is missing a sliver of curtilage, please resubmit. You resubmit. The portal says 'awaiting validation'. The client emails: any update?

**Planning portal admin is mostly monitoring and communication.** An AI agent cannot make the planning officer work faster, but it can check the portal status twice a day and email you when something changes. It can draft holding emails to clients that explain where the application sits in the process without you having to remember which stage it is actually at. It can log the history of submissions and officer comments so when the client asks why this is taking eight weeks, you have a timeline that proves it is not your fault.

Some practices use an agent that monitors the planning portal for multiple live applications, flags when an officer requests more information, and drafts a response based on similar requests from past projects. If the officer wants a heritage statement and you wrote one for a similar conservation area project last year, the agent pulls that statement, adjusts it for the new site, and drops it in a draft email for you to review and send. The officer gets a coherent response in 48 hours instead of two weeks, which measurably improves your approval rate because it signals you are responsive.

This is not complex AI. It is a scheduled task that scrapes the portal, a lookup against your project files, and a drafting agent that writes in your voice. The value is not in the sophistication. The value is in the 15 minutes you do not spend checking the portal every morning and the client relationship you do not damage because you forgot to update them for a month.

Client communications: keeping people informed without drowning in email

A residential client emails: when will the structural engineer send the calculations? You forward the email to the structural engineer. The engineer replies three days later. You forward that reply to the client. The client replies: can we move the steel beam 300mm south? You open Revit, check if that affects the floor joists, email the engineer again, wait for a response, then email the client. This is your Thursday.

**Client communication in architecture is mostly triage and forwarding.** An AI agent can monitor your project inbox, tag emails by urgency and type, draft responses to common questions, and route technical queries to the right consultant without you reading every message. It does not pretend to be you. It drafts a reply and puts it in your outbox for approval. You still send the email. You just do not write it from scratch.

The agent learns your tone by training on past client correspondence. If you always sign off with 'Best regards' and tend to explain planning delays by referencing council workload, the agent mirrors that style. If you have answered the same question about U-values five times this year, the agent recognises the pattern and drafts a reply based on your previous explanations. You tweak it, hit send, and move on.

Practices report that this cuts client email time by about 50 per cent, not because they are sending fewer emails but because they are spending less time per email. The client experience improves because responses are faster and more consistent. You are less likely to forget to update someone because the agent flags unanswered client emails after 48 hours. It is not about automation. It is about not losing track of a conversation because you were in a site meeting when the email arrived.

Revit and ArchiCAD integration: pulling AI into your actual workflow

If the AI agent lives in a separate web dashboard you have to remember to open, you will not use it. Architecture practice AI only works if it plugs into the tools you already have open all day, which for most practices is Revit or ArchiCAD, Outlook or Gmail, and whatever project management tool you pretend to update.

**Revit integration lets an AI agent read your model and surface information without you exporting files.** An operations agent can scan your Revit model, extract the schedule of materials, cross-reference it against your NBS library, and flag missing specifications. It can pull room areas and element quantities into a fee tracking spreadsheet so you can see if the project is still within scope as the design evolves. It can monitor when someone changes a key dimension and notify the relevant consultants without you remembering to email them.

ArchiCAD integration works the same way. The agent connects via the API, reads your project data, and pushes updates to wherever you track project progress. If you are still using spreadsheets to log consultant submissions and planning milestones, the agent can populate those sheets automatically by pulling dates from your project timeline and planning portal status. You stop maintaining two sources of truth because the agent syncs them.

The technical lift here is not trivial. Revit and ArchiCAD APIs exist, but they require someone who knows how to write the integration and map your data schema to the agent's input format. This is where working with an agency like Klevere matters. You do not want to hire a developer to figure out the Revit API while also learning how to train an AI agent. You want someone who has done it before and can deliver a working integration in weeks, not quarters. See our /solutions/ai-agent-development page for how custom integrations are scoped and delivered.

ARB and RIBA compliance: what you can and cannot delegate

The Architects Registration Board is clear: you cannot delegate your professional responsibilities to software. If an AI agent drafts a structural spec and you send it without review, you are still liable if the specification is wrong and the building fails. RIBA's code of conduct says the same thing. AI for architects is a drafting and coordination tool, not a substitute for your professional judgement.

**The compliance boundary is simple.** An AI agent can draft documents, retrieve information, monitor tasks, and flag inconsistencies. It cannot approve a design, sign off on structural adequacy, certify compliance with Building Regulations, or take responsibility for a decision. You review everything the agent produces before it leaves your practice. The agent saves you time by doing the first 80 per cent of the work. You still do the last 20 per cent, which is the bit that requires your ARB registration.

Some practices worry that using AI will create liability risk if something goes wrong. The risk is the same as using a Word template or a Revit family you did not draw yourself. If you review the output and approve it, you own the decision. If you do not review it, you are negligent whether the document came from an AI agent or an overworked Part 2. The agent does not change your duty of care. It changes how much time you spend on each task.

RIBA does not prohibit AI use. The guidance is to make sure you understand how the tool works, keep records of what was produced and how it was reviewed, and maintain the same standard of care you would apply to any other drafting or coordination method. If you can explain your process to a client or an adjudicator, you are fine. If you cannot explain it because you do not understand what the agent did, you should not have used it.

How Klevere approaches AI for architectural practices

Klevere does not sell architects a plugin. We build custom agents for the specific tasks that cost your practice the most time. That starts with a free 30-minute AI audit where we map your current workflow, identify the repetitive tasks that are eating your week, and recommend which agents will deliver the fastest return. Most architecture practices get value from three agents: a fee proposal drafter, an NBS specification assistant, and a planning admin monitor. We build those first, measure the time saved, then expand into client comms and consultant coordination if the practice wants to go further.

**Our /ai-os/operations-agent handles the bulk of admin work** for architectural practices: document assembly, specification drafting, planning portal monitoring, and task routing. It integrates with Revit or ArchiCAD via API, connects to your email and project management tools, and learns your office templates and past project data. We train the agent on your won proposals, your spec library, and your client communication style so the output reads like your work, not generic boilerplate.

For practices that need deeper integration, we offer custom agent development through /solutions/ai-agent-development. That might be a Revit plugin that syncs model data to your fee tracking system, an agent that monitors your planning portal accounts across multiple local authorities, or a client communication agent that triages your inbox and drafts responses. Every Klevere engagement is scoped individually after the audit, so we only build what will actually get used.

We work with practices from solo architects to 20-person studios. The smallest engagement is usually one agent deployed over four weeks. The largest we have done in this sector is a six-agent system covering proposals, specifications, planning admin, client comms, consultant coordination, and fee tracking. The common factor is that every agent is designed around a task the practice was already doing manually. We do not introduce new process. We automate the process you have.

If you want to see whether AI is a fit for your practice, book a free audit at /contact. We will walk through your current workflow, identify where you are losing time, and give you a clear recommendation on what to build first and what to leave alone. No obligation, no sales pitch. Just a practical conversation about whether AI can give you back the hours you are currently spending on admin.

The architecture practice that uses AI is the one that wins the next project

The client brief arrives on Monday. They want a fee proposal by Wednesday. Practice A spends Tuesday evening copying an old proposal, adjusting the scope, and hoping it lands. Practice B runs the brief through an agent that drafts the proposal in ten minutes, spends 30 minutes editing it, and sends it Tuesday afternoon with a note saying they can start the feasibility study next week if the client wants to move quickly.

Practice B wins the project. Not because the proposal was better written. Because it arrived 24 hours earlier and signalled that this practice is responsive, organised, and ready to start. The client does not know an AI agent drafted the document. They just know Practice B seems like they have their act together.

AI for architects in 2026 is not about designing better buildings. It is about running a practice that does not lose projects because you were too buried in admin to reply on time, does not haemorrhage fee because you forgot to track scope creep, and does not burn out your associates because they are spending half their week writing emails and checking the planning portal. The practices that adopt AI are not more creative. They are just less exhausted, which turns out to be a significant competitive advantage when the client calls and you actually have time to answer the phone.