Innovation Matters
Essays on AI, developer tools, and thoughtful software.
What's volatile about your technology?
We can't know exactly how current a model's knowledge is, but technology providers can show agents which parts of their technology change frequently.
Your agent should start with a hypothesis, not a plan
Let agents use what they know, but verify assumptions that may have expired before those assumptions become a plan.
DRY might be wrong for agent documentation
We normalized documentation to make it easier to maintain. Agents change the economics. Maybe the source should stay normalized while published workflows become self-contained.
I was wrong: there are bad prompts
A prompt that produces undesirable results is easy to call bad. But there's another kind that's harder to spot.
Your agents should learn for the organization
Agents learn while doing work, but most of what they discover disappears with the conversation. Organizations need a way to turn those discoveries into verified, current knowledge.