AI made every hiring claim cheap. A resume, a test score, a confident interview answer: all of them can now be produced in seconds by someone who cannot do the job. Tendent replaces claims with proof. Candidates do real work built for the job, then defend it live. The work and the defense are scored apart, and a person makes the call.
What broke
Every signal hiring runs on used to be expensive to fake. A strong resume took experience. A passed test took knowledge. A good answer took understanding. Now all three are free, and checking them still costs a skilled person’s time. That imbalance is the whole problem.
The result is a loop. Companies cannot assess at volume, so strong candidates cannot stand out, so candidates apply more with AI, so volume grows, so assessment gets worse. Each turn of the loop makes the next one harder.
The tools built for the old world do not stop it. Resume filters rank claims. Test libraries leak. Proctoring is beaten by a second device. AI interviewers score how well someone talks about a skill, which is exactly what a language model is best at. And AI detectors get worse as models get better.
The idea: you cannot defend work you did not do
Stop trying to catch AI. Start measuring what a person can actually do, and whether they understand it.
Tendent works in three steps.
The work score and the defense score are never mixed into one number. That is where the signal lives. Work 91 and Defense 89 is a strong hire. Work 94 and Defense 52 is a conversation you need to have. One blended score would hide the difference.
AI use is a skill, not a crime. The Defense asks how the candidate used it: what they asked for, what they accepted, what they threw out, and why. A candidate who directed a model well and can explain every line has shown something the job needs. A candidate who pasted an answer cannot explain it. Same mechanism, no detector, no accusation.
And a person makes the call. Every score comes with a reason a hiring manager can check, change, and teach the AI from.
Why it works
None of this is a guess. Each part of the design rests on published research.
Put simply: the research says test the real work, then interview about that work, and never trust a detector. That is the whole design.
What we have seen so far
Early results from evaluations run with design-partner customers in real hiring processes:
Want the full picture?
The detailed paper covers what this one leaves out:
- the ten requirements any modern skill evaluation has to meet, and how common tools score against them
- Solve, Defend and Verify in detail, including how evaluations are generated from the job and how the Defense chooses its questions
- identity and continuity: how Tendent builds confidence that the same person did all of it
- how the system learns from hiring managers and real hiring outcomes
- how the design maps to the EU AI Act, and what it does not claim
Request the full whitepaper: victor@tendent.ai
