THE RESEARCH / OUR INTERPRETATION
BNA’s August 2026 update reports widespread AI adoption alongside concern about cybersecurity and data quality. RIBA’s 2026 survey also records growing use. Neither finding establishes the maturity of an individual studio.
Start with a handover, not a tool list
Take one recurring task: preparing an interior concept review after a client changes the budget. Ask a designer to demonstrate the actual sequence, from the latest brief to the presentation. Where is the approved budget recorded? Which references may be uploaded? Can a colleague identify why one direction was selected? A smooth demonstration by one enthusiastic user is a starting point, not evidence that the team can repeat it.
- 01Skills
- 02Project context
- 03Tool connections
- 04Governance
Assess each dimension against one real workflow. A strong result in one does not compensate for a missing control in another.
What the assessment would reveal in a 24-person studio
Consider a proposed assessment for a 24-person architecture and interiors studio with six people using AI regularly. Its strongest user can produce a concept deck quickly, but the brief lives in email, the material references sit in personal folders and the project architect reviews the final PDF. The bottleneck is the handover between exploration and a defensible design decision. During the assessment, we would follow one recent change through that chain: who accepted it, which source superseded which, and where the design implications were checked. A useful finding might be that image skills are adequate while revision control is weak. The first investment would then be a shared brief register and review protocol, rather than another image subscription. This is an illustrative diagnosis; the method starts with the studio’s actual records.
Build a profile across four dimensions
We assess skills, project context, tool connections and governance separately. Image exploration might sit at Apply, while project memory remains at Explore and a single document workflow reaches Integrate. Record an observable example for each judgement. Avoid averaging these into a score that hides a missing permission rule or an undocumented source. The useful result is a map of dependencies: what must be resolved before this particular workflow can progress?
Evidence that separates Explore, Apply, Integrate and Evolve
For this workflow, Explore means a person can investigate a task with approved inputs and identify unreliable output. Apply means another colleague can repeat a documented method against the same acceptance criteria. Integrate adds maintained connections to project information, identifiable owners and a managed change process. Evolve adds coordinated specialist work with enforced permissions, monitoring and recovery. These are working interpretations for our assessment, not a certification scheme. We would ask to see a source pack, a completed run, an exception and a handover. A slide showing an agent diagram establishes none of these. Equally, a carefully maintained retrieval workflow may be more valuable than an elaborate multi-agent system. The relevant question is whether the operating arrangement matches the task’s consequence, frequency and variability.
| Activity | Current method | Proposed pilot |
|---|---|---|
| Information / context | 75 | 35 |
| Production | 105 | 45 |
| Review / correction | 60 | 75 |
| Total per run | 240 | 155 |
85 minutes net saving. A 20-hour setup takes approximately 15 comparable runs to recover, excluding licences and maintenance. Hypothetical example.
Choose a pilot with a measurable finish
For the concept-review example, the pilot could deliver a current context pack, three reviewed material directions and a decision log. Agree who owns each input, who reviews the output and what success means before starting. Compare total preparation time, correction effort and traceability with the current method. Include training and review time; faster image generation alone does not demonstrate a faster project.
Build the business case around review effort
Suppose the current concept-review task takes 240 minutes: 75 gathering information, 105 producing material and 60 reviewing it. An illustrative pilot takes 155 minutes: 35 preparing context, 45 producing candidates and 75 reviewing and correcting them. The net saving is 85 minutes, even though production alone became 60 minutes faster. If setup and training require 20 hours, the time-only break-even is roughly 15 comparable runs, before licences and maintenance. If review instead rises to 160 minutes, the saving disappears. These numbers are a worked calculation, not reported performance. The assessment should also track source traceability, rejected output and whether the work shifts from junior staff to scarce senior reviewers. A workflow that saves drawing time but consumes the project architect’s attention may have poor economics.
A practical 90-day sequence
Use the opening weeks to establish the baseline and information boundaries. Run the same bounded task more than once, including a changed brief. Then ask a second colleague to repeat it from the documentation. Expand only after the team can explain the result, correct an error and maintain the method. This is our proposed assessment approach, not a claim that one timetable suits every studio.
The workshop should end with an operating decision
A half-day assessment cannot validate the security of every platform or prove the reliability of a studio-wide integration. Its useful output is narrower: a workflow map, an evidence-based maturity profile, a prioritised pilot and a named owner. For the concept-review example, the pilot charter would identify the project revision, permitted source folders, the deliverable, the reviewer and a stop condition. It would also list dependencies that remain unresolved, such as access to the BIM environment or permission to upload client imagery. The follow-up is a go, revise or stop decision after repeated trials. Choosing to keep a low-volume task manual is a legitimate result. Maturity includes knowing where an integration will cost more to maintain than the work it removes.