How we vet agencies

Your software is only as good as the team that builds it. So we vet the team.

Agencies go through a four-stage gauntlet before they get to bid on your brief. The bar is the bar — we re-test every 90 days. Failure rate is high, and that’s the point.

~50
agencies applied
12
currently in the network
76%
rejection rate

The four-stage gauntlet

We don’t care about portfolios, awards, or how long an agency has been around. We care that they ship working AI-native software, fast, and that their workflows are actually built around modern tools. So we test for that. Four stages, all real work.

1

Build test

We send the agency a real brief — a problem from a past project, with anonymized data. They have 5 to 7 days to ship working code. Not slides, not a Figma file. A repo we can clone and run.

Output: a deployable prototype
2

Code review

A senior reviewer reads every commit. Architecture choices, security boundaries, prompt design, data flow, error handling, test coverage. They sign off — or the agency does not pass to stage 3.

Output: a written review · pass/fail
3

AI workflow verification

Agencies screen-share their daily workflow. Claude, Codex, Gemini, Cursor in active use — not just installed, used. If AI tooling is on the marketing page but not in the daily commits, the agency is out.

Output: a 30-minute screenshare review
4

Quarterly re-review

Every 90 days we re-test. The bar moves with the tools — an agency that was AI-first in Q1 may not be by Q3. Stale workflows, missed updates, slipping shipped quality — back to stage 1, or out.

Output: continued network access · or removal

Tools we expect to see in daily use

In 2026, an agency that builds without modern AI tooling is shipping at 1× speed while the rest of the world ships at 4×. That gap shows up in your timeline, your bill, and the polish of the final product. So this is non-negotiable: agencies in our network use these tools every day, and we verify it.

Claude Code
Code · review · refactor
Codex CLI
Pair programming · deep work
Gemini
Multimodal · long context
Grok
Realtime data · research
Cursor / Windsurf
IDE-integrated AI
v0 / Replit Agent
UI scaffolds · rapid spikes
If an agency still says “AI is just a fad” — they didn’t make it in.

What gets an agency rejected

  • Portfolio without code. Polished case studies are easy to make. Working repos aren’t.
  • Legacy stack with AI sprinkled on top. A WordPress shop that just discovered Claude isn’t AI-first — they’re a WordPress shop.
  • No senior engineer who can lead architecture. AI tools amplify good engineers; they don’t replace them.
  • No shipped products. Hackathon demos and side projects don’t count. We need to see software running in production with real users.
  • Slow on the build test. If an agency takes 14 days for what should be a 7-day build, every project they touch will overrun.
  • Defensive about code review. Senior engineers welcome scrutiny. Junior teams pretending to be senior get defensive. Tells.

What this means for your project

  • Three agencies bid on every brief, all of whom have proven they can ship working code, not just promise it.
  • The agency you pick has been senior-code-reviewed in the last 90 days — their architecture isn’t stale.
  • You’re never matched with an agency that has a portfolio but no production track record.
  • If an agency drops in quality, they’re removed from the network within 90 days. Your project team is current, not legacy.
  • You don’t do the vetting work. You don’t need to know what “AI-first” actually means under the hood. We’ve already done it.

Vetted agencies. Three bids. Working code in 5–7 days.

The whole point of this is so you don’t have to vet anyone yourself. Send us your brief, get three real prototypes back.

Post your idea