Smart, senior, great at architecture, and somehow slow and off with AI. ai-fluent shows you how candidates actually work with their agents, before you hire them.
there's a better signal. here's how ↓— Batman Begins, 2005
Evaluate your candidates by what they do, not by the result their AI produced. ai-fluent looks at the interaction between candidates and their agents, and evaluates the traits you care about.
Sessions work like this: tokens go into Redis with a TTL and refresh on every request. Want me to go deeper on the eviction policy, or on the refresh race condition?
yes, go deeper on both
That's enough on sessions, no business need yet. Let's move to the login flow.
Conversation steeringHere are 14 user stories for onboarding, covering SSO, team invites, role permissions and an analytics dashboard. Ready to push them to Jira?
let's go
Cut it to the five that get a new user to first value. And story 6: invites by email or by link? Those are different flows.
Preference for simplicityAccounts with fewer than 3 logins a week churn 4x more. I recommend a re-engagement email to all of them this week.
great, draft the email
Is it the logins, or are those just the small accounts? Split by plan size before we email anyone.
Critical thinkingWhat we measure
Here is what we measure. Six habits that hold in any role, and you can add requirements of your own.
Sample profile
Holds the wheel. Takes what's useful and moves on before the model drifts.
Reads the answer like a skeptic. Catches the wrong number before it ships.
States a view, invites a challenge, changes course when the argument is good.
Breaks the job into steps instead of one giant prompt and a prayer.
Says no to the extra layer, the extra library, the extra twelve paragraphs.
Looks past the happy path. Brings up edge cases before production does.
The problem
One teammate on autopilot slows down the whole team: the PR nobody asked for, the spec that misses the real question, the reply that promises a feature you don't have. Someone else cleans it up.
Slop isn't free. The people who clean it up pay for it.
A short exercise built for the role. Developers use their own coding agent. Product and CS work in a guided chat.
A wrong fact. A bad assumption. A tempting tangent. A plan twice as big as it needs to be.
Did they catch it, push back, or wave it through? A ten-minute review, with every claim tied to the transcript.
The report
Each criterion you chose comes with the exact moments behind it. You read, you judge. No black-box number deciding for you.
Built for the hiring manager's ten minutes between meetings.
Who it's for
See the Claude Code session, not just the diff. Who read the plan before approving it, who caught the bug the agent planted.
Find the PM who trims fourteen stories to five, and asks which flow you meant before writing either.
Hire the CSM who checks the AI's churn theory before it reaches a customer's inbox.
We're running early pilots with R&D and product teams in Israel.
Book a demo