Asked a targeted clarification
The prompt named the suspected failure mode instead of requesting a complete solution.
Evidence: Prompt 01 · 02:14
For engineering teams hiring in the AI era
See how candidates reason with AI, not just what they submit.
Run realistic AI-enabled assessments and review how candidates prompt, test, revise, and make decisions, not only the code they submit.
No credit cardCandidates need no accountNo screen recording
Candidate workspace Captured activity
Candidate workspace
42:18 remainingrate-limiter.tstests.ts
17private refill(bucket: Bucket) {
18 const elapsed = now - bucket.updatedAt;
19 bucket.tokens = Math.min(
20 bucket.capacity,
21 bucket.tokens + elapsed * bucket.rate
22 );
23}
8 tests passedRun 04
AI assistant
What happens when elapsed time refills above capacity?
Clamp the result to the bucket capacity before updating the timestamp.
I'll add the clamp, then rerun the burst tests.
The missing context
The final answer hides the most useful signal.
AI can polish an underprepared answer or help a strong candidate move faster. The submission hides the difference. DevTrace shows the process.
01
Did the candidate decompose the problem before asking AI?
02
Did they verify and test AI-generated suggestions?
03
Did they adapt when a suggestion was incomplete or wrong?
Example candidate report
Review the evidence, not a black-box score.
DevTrace uses AI to summarize observable session evidence. Every observation links to the underlying prompts, edits, tests, and revisions. Your hiring team reviews the evidence and makes the final decision.
The report describes recorded workflow activity. It does not claim to read a candidate's thoughts or replace human review.
Candidate 014 · Backend engineer
Token bucket rate limiter · TypeScript · 47 min
Final submission4 prompts6 runs1 paste
Evidence-linked observations
Review against the linked session activity
Tested the suggestion against edge cases
The candidate ran the burst tests, saw the failure, and changed the implementation.
Evidence: Runs 02, 03, and 04 · Edits 17 through 23
Revised the approach after a failed run
The final code differed from the pasted suggestion and addressed the failing assertion.
Evidence: Paste 01 · Edit 23 · Run 04
Made the architectural decision independently
The per-key bucket structure was present before the first AI prompt.
Evidence: Initial edits · 00:31 to 01:46
Suggested follow-up
Explore why the candidate changed timestamp handling after the failed burst test.
Edit and run timeline
Select an event in the full report to inspect its snapshot.
02:14Prompt 01Asked whether burst refills could exceed capacity.
06:48Run 02Two burst-limit assertions failed.
07:20Paste 01Inserted a 96-character clamp suggested by AI.
11:32Edit 23Reworked the refill expression and timestamp update.
14:05Run 04All eight tests passed after the revision.
Final submission
rate-limiter.ts · 61 lines · 8/8 assessment tests passing
DevTrace summarizes observable workflow signals. Your hiring team makes the decision.
View example reportHow it works
From task brief to evidence-backed review.
- 01
Create a realistic task
Use your brief, a template, or an AI draft.
- 02
Send a private link
Invite candidates directly. No account required.
- 03
Candidate works with AI
Solve in an editor with tests and an AI assistant.
- 04
Review process evidence
Review the submission with prompts, edits, runs, and revisions.
No extension installationCode and written tasksJavaScript, TypeScript, Python, C, and C++
Candidate experience
Signal without surveillance.
Candidates see what DevTrace captures before starting. It records activity inside the assessment and when the tab loses focus, never activity in other apps or tabs.
No webcam recording
No microphone recording
No screen recording
No hidden monitoring
Reports are limited to the hiring team that created the role. DevTrace does not claim certifications or a retention policy it has not formalized.
A different assessment model
Keep the flexibility of a take-home. Add the context of a working session.
| What reviewers can evaluate | Traditional interview | Unmonitored take-home | DevTrace assessment |
|---|---|---|---|
| AI use is allowed | Varies | Unclear | Yes |
| Candidate process is visible | Live only | No | Yes |
| Observations link to evidence | Interviewer notes | No | Yes |
| Candidate works asynchronously | No | Yes | Yes |
| Process and output reviewed together | Partial | No | Yes |
Testimonials
What people are saying.
“I want to know how someone got to the answer, not just whether they got there. DevTrace gives you that context.”
“People are already using AI at work. It makes more sense to see how they use it than pretend they will not.”
Design partner program
Run your next engineering role with us.
We’ll help create the assessment, onboard candidates, and review the first sessions. The free pilot covers up to three candidates.
Pricing
Start with a real hiring role.
The founder-assisted pilot covers your first three candidates.
Compare paid plansFree pilot
For teams evaluating DevTrace on a live role.
$0
- One active role
- Up to three candidates
- Full workflow capture
- Evidence-backed candidate report
- Private candidate links
- Founder-assisted setup
FAQ
Clear answers for hiring teams.
Is DevTrace a proctoring tool?
No. DevTrace does not record webcam, microphone, or screen activity. It captures workflow events created inside the assessment and notes when that browser tab loses focus.
What activity does DevTrace capture?
Prompts and responses from the built-in AI assistant, edits, code runs, large paste events, task timing, tab focus changes, the final submission, and the candidate’s reflection.
Can candidates use external AI tools?
DevTrace provides an AI assistant inside the workspace. If your process allows external tools, DevTrace cannot capture those outside conversations. It can only show workspace paste and focus change events.
How are evaluation observations generated?
AI summarizes session telemetry, the embedded AI conversation, test runs, submission, and reflection. Reviewers can inspect the underlying events instead of relying on the summary alone.
Does DevTrace make the hiring decision?
No. DevTrace organizes observable evidence and suggests follow-up questions. Your hiring team interprets that evidence and makes the decision.
What languages and task types are supported?
Coding tasks support JavaScript, TypeScript, Python, C, and C++. DevTrace also supports written and code review assessments.
How is candidate data handled?
Session data is available to the authenticated hiring team that owns the role. A formal public retention and deletion policy is not yet stated on this site; confirm your requirements with DevTrace before a rollout.
Does the candidate need an account or extension?
No. Each candidate opens a private assessment link in the browser. There is no DevTrace account or extension to install.