Durable Agent Harness
Dynamic Devices mark

Durable Agent Harness

Make the model powerful. Make the system trustworthy.

A public lab note from Dynamic Devices on the engineering journey of working with AI agents — clocks, stack, capability, and measurement honesty. No private life. No secrets dump.

Explore the stack View on GitHub

Tenure

Three clocks that matter

Not “first heard of ChatGPT”. These clocks mark when hands-on AI work, the agentic IDE, and a versioned harness each became real.

Explore

One place to walk the journey

Pick a lens. Stay as long as you like. Everything here is written to be safe on the public internet.

Measure

Felt speed is not a metric

We track tenure and capability. We refuse a marketing “N× more productive” number without a method.

What we count

  • Calendar tenure on the three clocks
  • Harness growth (skills / patterns) as investment
  • Outcomes that moved from impossible or uneconomic to routine
  • Proof-backed delivery — not lines of code

What we will not claim

  • A universal speedup versus every developer class
  • That agentic IDEs always win on familiar tickets (see METR 2025)
  • That more generated text equals more engineering
  • Anything that requires exposing private work to “prove” it

Publication bar

Public means redacted

This site is generated from a privacy-safe content pack. Private chats, family details, credentials, lab addresses, and competitive product plans stay out — on purpose. Read PRIVACY.md and run npm run privacy before publishing changes.