74% of UK architecture practices now use AI on most of their projects.

a year ago it was 59%. and 61% of respondents think AI will make it harder for early-career people to learn the job (that's you, by the way).

AI in architecture and construction has moved from "interesting demo" to "how the office works".

here's what's actually changing, and how to think about it as a student.

What's actually happening with AI in architecture and construction

four stories from this year that tell the whole picture:

  1. the profession has adopted it. RIBA's AI Report 2026 found 75% of respondents report better practice productivity. but only 17% say their designs are better because of AI, and 59% of practices expect staff reductions.
  2. the software is being rebuilt around it. at Autodesk University in September, Autodesk confirmed Forma will replace Revit over time, according to AEC Magazine. it also previewed an "agent-first" Autodesk Assistant expected in 2027, plus Autodesk Context, meant to capture WHY project decisions were made.
  3. robots are on site doing the admin. Tilbury Douglas put a humanoid robot called Douglas on a live site to capture 360 degree imagery and progress reports. the firm expects it to save about 40 hours a month (and it cost £15,000, less than a junior's salary).
  4. planning is next. the UK government launched AI planning tools in June. one prototype, tested with Barnet, Camden and Dorset, aims to cut householder decisions from eight weeks to four. another, Extract, digitises old planning documents and is estimated to save 250,000 hours a year.

notice the pattern? AI is taking the checking, logging and summarising first. the stuff juniors used to learn on.

What uni won't teach you about AI in architecture

architecture and civil engineering degrees are brilliant at design thinking and structures. where they lag:

→ software is taught as a fixed tool. you learn Revit like it'll be there forever. Autodesk just told the industry its replacement is coming. → the reasoning behind decisions isn't recorded. studio crits reward the final drawing. Autodesk building a product to capture decision context tells you where the value is moving. → site data is abstract. most students never see a progress report, let alone one a robot produced. → planning feels like a later problem. if AI starts drafting the first assessment of an application, the people who understand both design and policy get a lot more useful.

The critical-thinking move: 5 questions for any AI tool in AEC

use these on any AI design, site or planning tool you meet in studio, on placement or in a job interview:

  1. who carries the liability if it's wrong? a generated layout or an AI flagged defect still needs a named professional to sign it off. the UK planning prototype is explicit that a qualified officer reviews every assessment.
  2. what is it optimising, and is that what matters here? one AU demo generated more than 1,000 fascia options in a single workflow. the scarce skill is picking the right brief and the right criteria, not making option 437.
  3. what does it know about THIS site? soil, neighbours, local policy, the client's actual budget. a tool trained on general data can be confidently wrong about the one thing that matters.
  4. who owns the data and the workflow? AEC Magazine's own coverage flags that Autodesk, not its customers, decides what counts as "secret sauce" in training data, and that licensing could change if agents cut seat counts. ask that about every platform you learn.
  5. what skill am I skipping by using it? 61% of RIBA respondents worry about exactly this. if AI writes the site report, make sure you still know what a good one looks like.

What to do this month

  1. try Forma properly. Autodesk offers free education access. run a site analysis on a real plot near your uni and compare it with what you'd have concluded by hand.
  2. read the RIBA AI Report 2026. it's the best snapshot of how UK practices actually use AI, and you'll sound sharper than most people in a placement interview.
  3. do one task twice. produce a small massing study or site summary the slow way, then with AI. write down what the tool got wrong. that list is your value.
  4. read three delegated planning decisions on your council's planning portal. that's the exact work the government's prototype is trying to speed up, so understand it before the tool does it for you.

so TLDR: AI is eating the checking and logging work first. the students who understand liability, data and the brief will be the ones signing the drawings.

which part of your design process would you never let an AI do without checking it yourself?

Sources

Get the next issue by email

theVANTAGEd is written by students to bring critical thinking to AI across real industries. One issue every two weeks.