A sandboxed company staffed by autonomous AI teammates. People do the work; we score the replay — and the same scenario trains your team and screens your next hire.
TrainHirePractice
Eje David — Co-Founder & Technical LeadAdoyi Daniel — Co-Founder & Go-to-Marketcontact@finderk.com · finderk.com
The best predictor of performance is watching someone do the job.
Work sample (do the job).54
Structured interview.51
Unstructured interview.38
Résumé / experience.18
Yet almost no one uses work samples — they're expensive to build, impossible to standardize, and don't scale. So companies default to the weakest signals.
Predictive validity of selection methods — Schmidt & Hunter (1998); rank order confirmed by Sackett et al. (2022), J. Applied Psychology.
Describe the moment to train or test. An authoring agent proposes the company, the coworkers, the tools, the seed data and the rubric — then drops you into a prefilled editor. Authoring goes from weeks to minutes.
02
You pay only for the usage your content generates.
Per-session compute "drain" is covered inline from the buyer's payment, so creators pre-fund nothing and a buyer is never blocked. Finderk keeps the drain plus a 15% marketplace take.
Authoring scales supply. Usage-based pricing makes every session gross-margin-positive from day one.
AI got cheap enough to staff a company. The cost of capable AI falls ~10× per year — a given level halves every 6–8 months. Believable agentic coworkers became affordable to run at scale only in 2024–26. (a16z LLMflation · Epoch AI)
02
Work went digital, so the job is simulatable. Remote settled at ~25% of US workdays — a permanent ~5× shift. The job is now a set of apps you can replay and score. (Stanford WFH · Gallup)
03
Skills-based hiring is here — but the evidence isn't. 81% of employers now use skills-based hiring, yet most still lack a credible way to measure skill. (TestGorilla 2024 · Burning Glass)
Gartner: enterprise apps embedding task-specific AI agents go from <5% in 2025 to 40% in 2026. Finderk's core mechanic rides the fastest-growing category in software.
Full multi-app workplaceSingle skill / channelScripted role-playAutonomous AI agentsFinderkAnthroposMursionVervoe · CanditechCodeSignalHyperbound · YoodliDeeptune (trains agents)
Closest analogue Anthropos ($2.7M pre-seed) does dual hire+train, but with scripted personas & task building-blocks — not autonomous agents in a live, integrated workspace. Coding tests, sales role-play and static work-samples each cover one axis only.
Coworkers that behave like real teammates — open-ended, reactive, never a branching role-play tree. The bar competitors haven't cleared.
One artifact, two budgets
The same simulation trains and hires — twice the wedge, and a proprietary dataset of how people actually perform the work.
AI-authored marketplace
Talk-to-build authoring + a two-sided catalog → compounding content and network effects no single-vendor tool can match.
Usage-based economics
Every session pays for itself; pricing scales with value delivered, not seats sold.
The category is forming around us — Mercor $10B, Workday's ~$1.1B Sana acquisition, top-tier funds backing AI that works and learns. No one has fused it into coworkers you practice and get hired alongside.
On a SAFE — ~18 months of runway to turn a working product into paid pilots, validated unit economics, and a live marketplace.
~50% · BUILD
Product & engineering
Reliability, scale, and the unit economics that make usage-based pricing work; ship great authoring + the marketplace.
~15% · RUN
Compute & infrastructure
Run many sessions at once, affordably — the cost base behind the margin story.
~35% · SELL
GTM, pilots & hiring
Convert hiring & L&D design partners into paid pilots and prove the dual-use ROI.
Milestones (12–18 mo): first paid B2B pilots across 2–3 beachhead roles · marketplace live with external creators · validated per-session unit economics.