Start with the contribution
A portfolio page can show an attractive AI interface while saying little about the designer’s role. Ask which users the team studied, which workflows it changed and which deliverables it owned. Product strategy, visual design, usability testing and frontend implementation can be separate pieces of an engagement.
The directory’s design shortlist includes specialist studios and broader product teams. Read the linked account for each case. When the source credits the client’s engineers with implementation, retain that distinction in your evaluation. An agency should not need to claim model ownership to demonstrate useful design experience.
Inspect the difficult moments
Ask to walk through an incorrect answer, an incomplete result and a user who cannot tell whether to trust the output. Useful evidence includes the alternatives explored and the reasons for the chosen interaction. A polished happy path alone leaves these questions open.
Look at what happens before and after generation: entering useful context, setting expectations, checking source material, editing a result and deciding whether to act. The importance of each moment depends on the task and consequences. Ask for examples close to your own workflow rather than expecting one interface pattern to suit every product.
- Can people tell what the system is doing?
- Can they correct or reject a result?
- Are consequential actions distinguishable from draft suggestions?
- What happens when the system cannot complete the task?
Ask how the design was evaluated
Request the research question, participant context and the stage at which findings changed the design. A prototype can reveal interaction problems, but its results do not establish a production system’s reliability or adoption. Ask what remains to be tested with real data and real operating conditions.
Microsoft’s HAX guidance organizes human-AI interaction across initial use, ongoing interaction, mistakes and change over time. It provides a useful primary reference for a review conversation. It is not evidence that a particular provider follows it or that a project is certified.
Make the next team’s needs explicit
Agree how design decisions, states, components and unanswered questions will be handed to engineering. Identify who owns evaluation inputs and who reviews changes when model behavior shifts. If the design team also implements the interface, ask where its responsibility ends and backend ownership begins.
A useful selection conversation ends with a proposed research and delivery scope tailored to your product. It does not need a claim that a studio is universally “best” at AI. Use the case evidence to choose who is worth that conversation.
Sources and editorial context
Procurement checklists are Point of AI’s editorial guidance. The following primary sources support the specific principles identified here; they do not endorse this directory or its listed providers.
- Microsoft HAX: Guidelines for Human-AI Interaction ↗
Interaction guidance covering first use, ordinary interaction, mistakes and use over time. Source reviewed 8 November 2026.