$1,000 Competition!
Select your conference to learn more. October 3rd deadline.
See how an engineer really works with AI, before you hire.
Agility Sonar is a one-hour technical assessment for hiring teams. Candidates work through a real, unfamiliar codebase using AI tools, and you get a clear, evidence-backed report on how well they actually use AI to understand, verify, and fix code, so you can evaluate this new generation of AI-assisted engineers before you hire them.
When a line item is deleted, where does the invoice total get recalculated?
DELETE /invoices/:id/items removes the row and returns 204. No recalculation is triggered on that path; totals are only recomputed on item create.
Why Sonar
Interviews test the wrong thing. Sonar tests how engineers actually work with AI.
Your engineers direct AI every day. Coding puzzles measure typing an algorithm from memory. The gap between what interviews test and what the job is has never been wider.
- Sonar measures the actual skill: scoping, prioritization, verification, and depth when auditing unfamiliar code with an AI at hand.
- The AI never gives the audit away. The built-in assistant is deliberately passive: it answers factual questions but never volunteers findings or ranks severity. Direction has to come from the candidate.
- AI use is the exercise, not a threat to it. There is nothing to police, because directing the AI well is exactly what's being graded.
How it works
Send one link. Get back a graded report on how they used AI.
Send a session link
No accounts, no setup for the candidate. They open the link, read the brief, and press Start. A permanent /practice link lets them warm up on a mock codebase beforehand.
They audit by directing AI
A read-only, full-stack codebase with planted issues across API, frontend, and deploy surface. A question budget and a clock. The candidate probes, verifies in the file browser, and logs findings.
You get a deterministic grade
An ensemble of AI passes extracts the facts; Python computes the score. Same session, same grade, every time. A styled PDF and a login-free share link go straight to the hiring panel.
What's inside
A complete platform for running and grading AI-based interviews, not just a chatbot bolted onto a coding test.
Every question traced
The full transcript, file opens, and timing are recorded and correlated. The report shows which questions did the work, quoted in the candidate's own words.
Claims verified against the code
Candidates commit to findings in a graded log. At report time a dedicated pass checks every claim against the actual codebase: supported, unsupported, or incorrect.
Deterministic grading
AI extracts facts; a fixed, versioned Python formula computes the 0-100 score. No vibes, no drift, no "the model felt generous today".
Watch it live
A real-time observer view per session: transcript, activity feed, progress checks against the answer key, and the ability to nudge or send priced suggestions.
Compare whole cohorts
One page, every candidate side by side: a discovery grid of every planted issue against every candidate, unaided discovery, question quality, and severity calibration.
Integrity signals, not lockdowns
Tab-aways, pastes, and large copies are recorded and disclosed, never blocked. Prompt-injection in candidate text is structurally neutralized before it reaches a grader.
The report
A detailed report on every candidate, and how they stack up.
See exactly where a candidate ranks against your own pipeline and against the top engineers around the world.
- Percentile benchmark scored the same way for every engineer who takes the exercise, so the comparison holds regardless of background or location.
- Ranked influential questions with what each one surfaced and why it mattered.
- Coverage map of evaluation areas reached vs. missed, severity-weighted.
- Severity calibration: did they call a critical a critical, or call everything critical?
- Seniority read broken into verification, scoping, calibration, and breadth.
GRADE-V7 · ENSEMBLE 3/3 · DETERMINISTIC
Built to be fair
A score you can explain and defend to any candidate.
Hidden answer key
The planted-issue manifest is never served to the candidate and never enters the assistant's context. No prompt injection can leak it.
Human-fixed penalties
Hints carry a grade discount an interviewer approved and froze at send time. The AI proposes; only a human-set number ever touches the score.
Every assist disclosed
Nudges, suggestions, and AI-drafted findings are all flagged on the report, so a reader always knows what the candidate did alone.
Comparable by construction
Exercise versioning and cohort presets record what each candidate actually sat, so you never compare scores from different tests.
Use cases
Learn how teams like yours use Sonar.
Senior engineering hires
HiringAgencies & consultancies
Bench qualityHigh-volume screening
ScaleInternal calibration
TeamsTrusted by
Backed by the team hiring teams already trust.
Sonar comes from AgilityIO, the software partner behind engineering teams at companies like these.
Why Agility
Built by engineers who hire engineers.
Sonar isn't a side project. It's the same tool AgilityIO uses to grade its own engineering pipeline, so every rubric, leak guard, and fairness rule has been tested on real candidates before you ever see it.
Runs in your world
A single self-hostable container. Multi-org with per-org API keys, spend budgets, roles, and audit trails. Your candidates' data stays yours.
Used on real candidates
Sonar grades AgilityIO's own engineering pipeline. Every rubric change, leak guard, and fairness rule came from running live interviews, not a whiteboard.
White-label ready
Name, tagline, and assistant persona are configurable under the same flexible licensing model AgilityIO uses across its own products.
FAQ
Questions hiring teams usually ask.
How much does Sonar cost?
Pricing depends on your team's hiring volume, session needs, and whether you want white-labeling or self-hosting. Contact us and we'll help you find the right fit.
How long does a Sonar session take?
About one hour. Candidates get a single link, work through a real, unfamiliar codebase with an AI assistant at hand, and the graded report is ready right after.
Can the built-in AI assistant give away the answers?
No. It only answers factual questions about the code, and it never volunteers findings or ranks severity. Whatever gets surfaced has to come from the candidate directing it well.
How is the score calculated?
By a fixed, versioned formula, not a model's impression. The AI extracts facts from the session, and a deterministic scoring pass turns them into the 0-100 result, so a score means the same thing every time.
Can I compare candidates against each other?
Yes. Every candidate who takes the same exercise is easily compared to the others, and benchmarked against every other engineer who has taken the exercise on Sonar, so you can see how your pipeline stacks up against top talent around the world, not just against each other.
Is anything about the session hidden from me as the interviewer?
No, the opposite. You get a live observer view of every session, and a report that discloses every nudge, hint, or AI-drafted suggestion the candidate received.
Ready to get started with Sonar?
Tell us a bit about your team and we'll set up a live demo session you can sit yourself.