Critical Friends/

Critical Friends

How it works.

Tell Critical Friends what your users need to accomplish, and know whether they still can after every change. Crities, synthetic users with their own goals and situations, try those jobs in your product and your AI assistants in real browsers. What gets in their way becomes findings you can act on and retest.

Start a pilot

01 / Your product

Point it at your product.

Add the site, app, checkout or assistant you want tried, an address Critical Friends can reach, and the first job your users come to do.

  • Environment
  • Intent
TeamNest in the Critical Friends workspace with its reachable environment and first intent
Actual workspace screens · Fictional TeamNest study

02 / Personas

Choose who tries it.

Describe the people you build for in a sentence, and Critical Friends drafts the perspectives. Each Criti attempts the job as one of them.

  • Background
  • Goals
  • Decision style
  • Starting situation
Critical Friends Design / Personas

AI persona

Maya Chen

Busy team lead

Runs a small design studio. Comfortable with project tools, but rarely has time to explore settings. Brings contractors into projects as work comes in.

What she wants
Get a contractor working on one project without changing the whole team’s plan.
How she decides
Looks for familiar language and a clear next step. Checks the cost before committing.
What gets in her way
Limited time, an unpredictable contractor budget, and concern about sharing too much access.

Design the perspective

Give the feedback a point of view.

Describe the background, goals, habits, and concerns that shape how someone uses your product. Those details give a Criti reasons to explore, hesitate, or ask questions.

Then choose her starting situation.

Role & access
Workspace owner
Account state
Free plan · First contractor invitation
Device
Desktop browser

What this perspective helps you learn

Can someone who needs limited access for a contractor find the right option and understand the cost?

Example persona · Maya Chen

03 / Study

Choose what to keep covered.

A study groups the sessions worth running again. It explores several questions at once: can people finish their tasks, do the choices make sense, and where do they get stuck. Run it once to explore, then keep it covered as the product changes.

  • Study
Critical Friends Research / Studies

What we want to learn

Typed by you, or handed to your AI agent as one prompt.

PropertyTeamNestEnvironmentstaging.teamnest.example

Intent

Invite a contractor to one projectPulled from the question

Perspectives

  • Maya ChenBusy team lead · New owner · First visit
    Laptop
  • Arun ShahCautious founder · New owner · First visit
    Laptop
  • Sam RiveraReturning admin · Existing workspace · Returning user
    Phone

Sessions

3 plannedOne per perspective, run in parallel in a real browser

Study readyLaunch, or let your agent launch it

Illustrative planning sequence. Personas and situations come from the TeamNest example; the study screen shows the same plan in the workspace.

In the workspace

The same plan on the study screen: the question, the three perspectives, and their planned sessions, ready to prepare and launch.

Study detail: can a new owner invite a contractor without touching billing, with its planned sessions
Illustrative planning sequence · Study screen is an actual workspace screen

04 / Sessions

See every attempt.

Each session is one Criti attempting one job in a real browser. Play it back to see every hesitation beside the click that caused it.

  • Session
  • Playback
  • Keyframe
Critical Friends Evidence / Session
Maya’s first visitBusy team lead · Invite a contractor to one project
Illustrative AI session · No audio
Example app being testedTeamNest · Project management
teamnest.example / workspace
TeamNestAcme studio

Your workspace

A little teamwork goes a long way.

Keep your next launch moving with everyone in one place.

Website launch12 tasks · Just you, for now
00:00 · Looking over the workspace
00:00 / 00:27Quick playback 2.8s / momentKeyframe 1 / 5

Illustrative session with a fictional app and quotes.

Meanwhile

Three perspectives, three sessions. See how each Criti approaches the experience.

Maya

Busy team lead

Team
Workspace / Team
Your team
MMaya ChenWorkspace owner

1 of 1 seats used

Invite teammate

No seats available

Looking for the invitation

Arun

Workspace admin

Access
Workspace / Access
Choose a role
Member

Access to workspace projects

Project guest

Only the projects you share

Keep billing and settings private.

Checking contractor permissions

Sam

Returning teammate

Project
Website launch / People
Invite to project
contractor@example.com
Project guest Free

Access to this project only

Send invitation
Inviting from the project
Illustrative sessions · Three AI perspectives

05 / Observations

Read what they ran into.

Every hesitation and every “I wish” becomes an observation, linked to its moment in the session.

  • Observation
  • Friction
  • Ask
Observations screen listing frictions and asks from the TeamNest sessions
Actual workspace screens · Fictional TeamNest study

06 / Findings

See how the problem develops.

Your triage agent rolls related observations into findings, then tracks how often each one happens, when it was first and last seen, and whether it comes back after a change.

  • Frequency
  • First seen
  • Last seen
  • Regression
Critical Friends Workspace / Findings
Example · TeamNest, a fictional app

Finding · Contractor invitation

Inviting a contractor feels like a billing decision.

Retest clear · ready to close

New owners meet seat pricing before they understand guest access. It first appeared after seat pricing moved onto the team page, and the retest after the fix came back clear.

Sessions
111

of 414 before the fix

Crities
38

5 of 8 personas

First seen
Sep 9

after the seat pricing change

Last seen
Sep 22

before the fix

Frequency and health, run by run

Sessions that hit itClearSome sessionsMany sessions

9 runs · 474 sessions · 38 Crities · Staging · Counts reflect session coverage, not the share of real customers affected.

Sep 22 · Who hit it

26 of 64 sessions · 5 personas
Finding history concept · Illustrative study data

07 / Act & retest

Turn feedback into action. Then retest.

Pick the findings worth solving and make the change, or hand it to your AI agent. When you feel the problem is addressed, queue the finding for retest. Close it when the retest evidence shows it is resolved, or keep it open if the problem remains.

  • Finding
  • Queued for retest
  • Closed
Critical Friends Workspace / Findings
Example · TeamNest, a fictional app

Finding · Queued for retest

Inviting a contractor feels like a billing decision.

Change made: the invite screen now explains free guest access before seat pricing.

  1. 01 / Act on it

    Make the change, then queue a retest

    Make the change yourself or hand it to your agent. When you feel the problem is addressed, queue the finding to check the updated experience.

  2. 02 / Review the evidence

    Did the problem go away?

    New sessions try the same job. See whether the same friction shows up, and for whom.

  3. 03 / Resolve the finding

    Close it when resolved

    Keep the retest evidence with the finding. If the problem remains, keep the finding open for further work.

Example finding lifecycle

08 / Your agent

Your AI agent can run the whole loop.

Every screen is also reachable over MCP, so the agent you already use can run the whole loop, from setup to implementing fixes.

  • MCP
  • Setup prompt
Critical Friends Your AI agent / MCP
  1. Your productAdds your product and the address to test from your URL
  2. PersonasDrafts perspectives from a sentence about your users
  3. StudyPlans the sessions and runs them again when the product changes
  4. SessionsLaunches them and follows progress
  5. ObservationsReads every friction and ask as it lands
  6. FindingsDrafts candidates and brings you the call
  7. Act & retestImplements a fix, then retests the finding and shows whether the problem is resolved
Where you come in

You choose the jobs worth covering, decide which findings are worth acting on, and set the scope your agent may work within. Everything else can run without you.

Start a pilot How agents plug in
Every step is available as an MCP tool · You set the scope

Getting started

What to expect.

Do I need to design every persona or write test scripts?

No. Tell us about your product and the jobs people come to do, and we’ll propose the perspectives and situations to cover them. Crities explore through the visible cues in your product, without a prescribed sequence of clicks.

Can we use a local product or an experience behind login?

Use a public URL or a tunnel to a local or private environment. For signed-in experiences, your team provides test accounts with suitable permissions and realistic data. Your development agent can help prepare those accounts and connect the environment.

What happens to a finding over time?

Related observations build up under the finding. You can see how many sessions and personas encounter it, when it first appeared, and when it was last observed. When you feel the problem is addressed, queue it for retest. Close it when the evidence supports resolution; otherwise, keep it open.

What if I disagree with a finding?

You can defer it, mark it as noise or a misinterpretation, or explain why it conflicts with your strategy. Your reasoning stays with the finding and helps improve future triage.

How does this fit with user research?

Use Crities continuously and real users where they matter most. Crities surface friction and questions between research rounds, before real users meet each change. Real conversations and product data remain essential for understanding demand and validating broader conclusions.

What will they notice
in your product?

Start a pilot