AI agents can turn instructions into software. A working result can still be the wrong result.
We have come to value:
Human collaboration and accountable judgment over autonomous execution
Value demonstrated in use over software produced
Behavior verified against shared intent over instructions followed
Changing course when evidence changes over carrying out a plan
The things on the right matter when they serve the things on the left.
Twelve Principles
- Make a difference worth making. Meet a need or reveal a new possibility.
- Learn from real people. Simulations cannot establish demand.
- Deliver early. Build small, complete experiences that teach you something.
- Build less. Every feature, dependency, and agent must earn its place.
- Bound the investment. Set an appetite, limit human and machine costs, and protect quality.
- Work together around shared intent. Discuss uncertainty directly. Give people room to challenge decisions. Keep the resulting intent visible to teammates and agents.
- Control autonomy. Bound agents' authority and make consequential actions recoverable.
- Verify independently. Match the rigor of testing to the risk of being wrong.
- Measure what matters. Value and learning are progress. Generated output is not.
- Improve the whole flow. Account for review, recovery, and work in progress.
- Design for lasting use. Consider adoption, support, reliability, and economics.
- Follow the evidence. Revise, simplify, or stop when reality demands it.