Skill verification for the post-AI hiring stack
Anyone can submit a perfect answer. Show us the work.
Showwork runs a short adaptive session on a real task from the role and scores how a candidate actually thinks — including how well they use AI. The result is a portable skill profile, not another resume.
01 · Pick a task
Prompt
A production deploy passes CI but returns 502s. Walk through your first three checks and explain what evidence would change your direction.
Waiting for a response_
Demo scoring is simplified. Real sessions run 20–35 minutes.
Hiring signal fell off a cliff in eighteen months.
0x
growth in applications per opening since 2022
~0%
of job seekers now use AI somewhere in their application
0 sec
average time a recruiter spends on a resume
applications
302
index · 2022 = 100
usable signal
12%
relative strength
Both sides automated. The screen became noise on noise.
Figures based on published labor-market and recruiting industry reporting.
Two candidates. Identical on paper.
The artifact stopped separating people. The process still does.
Senior Product Analyst
Strategic, data-driven operator. Led cross-functional initiatives. SQL · Python · AI · Growth.
B.S. Economics · State University
Senior Product Analyst
Strategic, data-driven operator. Led cross-functional initiatives. SQL · Python · AI · Growth.
B.S. Economics · State University
Thirty minutes of real work, watched closely.
Calibrate
You describe the role. Showwork builds tasks from your actual work, not from a question bank.
Observe
The candidate works the task. AI tools are allowed and logged. The system probes with follow-ups when reasoning gets thin.
Score
Six dimensions, each with the specific moments in the session that produced the score.
Carry
The candidate keeps the profile and can send it to any employer. It expires in 12 months.
Six things a take-home can no longer tell you.
DECOMPOSITION
Breaking the ambiguous into the solvable
Does the candidate structure the problem before touching it?
VERIFICATION
Catching the confident wrong answer
When the model hallucinates mid-session, do they notice?
DIRECTION
Getting useful output from a model
Prompt quality, iteration, knowing when to stop delegating.
TRADEOFFS
Naming what gets sacrificed
Unprompted articulation of cost, risk and constraint.
RECOVERY
What happens after a wrong turn
Speed and cleanliness of backtracking under time pressure.
COMMUNICATION
Explaining the reasoning to someone else
Can a colleague act on the explanation without rework?
Faster screens, and screens that survive a reference check.
0 min
median session length
0%
candidate completion rate
0.0x
reduction in first-round interview time
0 mo
profile validity before recalibration
Metrics from design-partner cohorts.
Built to be audited.
Role-derived tasks
Every task traces to work the role actually does, which is the defensible basis for any selection tool.
Scores with evidence
Every dimension links to the session moments that produced it. Nothing is unexplainable.
Audit-ready by design
Structured logging and outcome tracking built for jurisdictions that require bias audits of automated employment decision tools, including New York City.
For candidates
You own the profile. Not us, not them.
Carry one verified profile across employers.
Use AI openly. What matters is how well you direct it.
Replace repeated take-homes with one focused session.