Don't just assess the answer.
See how people work with AI.
People do real tasks with Claude or GPT in an isolated sandbox, never your own code. SkillMark captures the prompts, the decisions, the checking, and the output, and turns them into an evidence-backed profile on the four axes of AI fluency. Use it to hire, and to see where your existing teams stand.
Invite-only access. Book a demo and we'll get your team set up.
The participant's workspace: task brief, code editor, and AI assistant side by side.
Demo session — the candidate and the data shown are fictional.
How It Works
One assessment in place of a multi-round process.
Your current process
- Phone screen / initial call
- Take-home assignment
- Review meeting with team
- Second round interviews
- Team debrief & decision
2–4 weeks
Time to decision
8+ hrs
Team hours
>$500
Per candidate
With SkillMark
Recommended- The person completes a role-specific task with AI
- Every prompt, edit, and decision captured
- Evidence-backed ratings on the four axes of AI fluency
- Tasks built to show whether someone asks the right questions
- Full session replay available
<30 min
Review time
30-60 min
Per candidate
Any
Role you hire for
4
fluency axes
Designing, Supervising, Reviewing, Ownership — SkillMark's names for Anthropic's AI Fluency competencies (Delegation, Description, Discernment, Diligence). Reported as four separate ratings, so a strong axis can't hide a weak one.
90+
ready-to-send assessments
Built and ready to send across every role you hire for — engineering, PM, marketing, design, QA, DevOps, data engineering, sales, CS, finance, legal, and more. Send one today, or author your own.
Full
session replay
Prompts, edits, and decisions on one timeline, recorded only after the participant consents.
The Evidence
Every rating points back to the session.
This is what a reviewer sees first: an overall band, the four axes with the moment behind each, and questions to follow up on.
Illustrative example
Specified the work well and verified it with tests. Accepted one flawed AI fix without checking it.
“Planned the parser split before prompting”
“Gave the failing test as the spec”
“Kept an off-by-one error from the AI”
“All tests green before submitting”
Follow-up questions
- Walk me through why you accepted the retry fix.
- What would you test next in the date parser?
Use them in an interview, or in a development conversation with a team member.
The Scoring Model
Four axes, not one number.
A single blended score lets a strong axis hide a weak one. The four axes come from Anthropic's AI Fluency framework; the evidence behind each rating comes from the session.
Fluency vector
Illustrative session
Designing is most readable next to Reviewing: heavy delegation with strong discernment is expert work; heavy delegation without it is not.
Features
The work behind the answer.
Traditional assessments grade the output. SkillMark shows how the person got there with AI.
Four axes, not one number
Designing (Delegation): choosing what to give the AI and what to keep. Supervising (Description): how clearly the work is specified. Reviewing (Discernment): whether AI output is checked and corrected. Ownership (Diligence): whether the result is shipped responsibly and owned. Built on Anthropic's AI Fluency framework.
Full session replay
Screen recording with event timeline overlay. See every keystroke, every prompt, every decision — reviewable at any speed.
Hire externally. Assess internally.
The same tasks and the same four axes, used on candidates you're considering or on the teams you already have. For internal teams you get a baseline, re-assessments over time, and a team view.
The review: evidence of how this person worked with AI, linked to the session.
Demo session — the candidate and the data shown are fictional.
Security & Privacy
Detailed evidence, collected responsibly.
SkillMark records a lot about how people work. That's why the limits are built in.
Isolated sandbox
Tasks run in an isolated, standardized environment. Participants never touch your code, your repositories, or your production systems.
Consent first
Before a session starts, participants see what is recorded and must agree to screen recording and activity capture.
Published retention windows
Session data is kept for defined periods and deleted on verified request.
Data retention →Known AI providers
AI assistance runs on Anthropic or OpenAI. Your organization chooses which one.
Read our Privacy · Terms · Data Retention
FAQ
Common questions
Still have questions? See all questions →
See how people actually work with AI.
Every prompt, decision, and output, before you make an offer or plan your team's next step.
Invite-only access. Book a demo to get started.
The Films
The argument, in eight pieces.
Short films on what AI-era hiring actually misses — and one long essay on the eight-hundred-and-eighty seconds where it all goes wrong. Swipe through.