Design partners · 3 slots

Put your agent on a time machine. Free, for 8 weeks.

I'm looking for a small number of teams running LLM agents in production to work with directly. You get a weekly drift report on your actual agent; I get the thing no incumbent has — real operator feedback from the config-rollback trenches.

The offer, in full

What you give, what you get.

No hidden asks. This is the entire deal — the same text I send in email.

YOU GIVE

  • 30 minutes a week — one feedback call, that's the whole time commitment after onboarding.
  • Under 30 minutes to onboard — I do the integration work myself; you review it.
  • If the reports prove useful: a quotable sentence and logo permission. Only on success — never before.

YOU GET

  • A weekly drift / diff report on your actual agent — which configs changed, which behaviors moved, which change is the suspect.
  • A config time machine — every state of your agent's prompt, tools, retrieval, model, and KB, versioned and diffable.
  • A direct line to the person building it — your stack decides the roadmap's first adapter, literally.

The trust terms, up front

Local-first: your config never leaves your machine. You run the CLI; secrets are redacted at capture; you share only the report. Mutual NDA available before we discuss anything. No telemetry; delete-on-request. Full details on the security page.

How the 8 weeks run

With a pre-agreed definition of success.

We define the activation event at kickoff and check it honestly: the report surfaced at least one change or drift you didn't already know about — and you did something in response. If that never happens, the partnership didn't work, and I'd rather both of us know.

WEEK 0

Onboard

Config extracted, eval set drafted from your real cases, first snapshot taken. I do the work; you review it. Target: under 30 minutes of your time.

WEEK 2

Activation check

Has a report told you something you didn't know yet? If not, we look at why — usually the eval surface, sometimes the tool.

WEEK 4

Review

Mid-point honesty: is the weekly report earning its slot in your inbox? Scope adjustments happen here.

WEEK 8

Decide

Continue as a paid pilot (design partners get 50% off year one), or walk away with your snapshots and reports — they're yours either way.

Fit

This is for a specific kind of team.

Three slots exist because I do the integration personally and cap my time per partner. So let's not waste yours:

A GREAT FIT IF

  • You run at least one LLM agent in production — real users, real consequences, not a prototype.
  • Your eng/ML team is roughly 2–20 people — big enough to feel the pain, small enough not to build everything in-house.
  • You've felt a silent regression — you can name the incident — or you manually revert configs today.
  • Someone owns reliability: an ML platform lead, head of AI engineering, or the senior engineer everyone actually asks.

NOT A FIT IF

  • Your agents are demo-stage — come back when they meet users; it'll be cheaper to start clean anyway.
  • You version everything in MLflow and it genuinely covers prompt + tools + retrieval + KB together — you've built it; I'd honestly love to hear what it cost.
  • You need a compliance certificate more than an answer — agentsnap is an early-stage tool with a local-first architecture, not a certified platform (details).

Also a strong fit: AI agencies and consultancies running many client agents — one tool, N configs, and "reproduce the state a client's agent was in last week" stops being an archaeology project.

Apply in one email.

A few lines about your agent, your stack, and the last time its behavior surprised you. I reply to every application personally — usually within a day.