Thinking on software identity, agentic development, and the engineering decisions that persist.
The sharpest objection to identity engineering is that you can't formally verify code against natural-language constraints. That's true. It's also irrelevant. Every quality mechanism in software is probabilistic. Reconciliation just needs to be better than nothing.
Specs drifted. Docs rotted. Tribal knowledge walked out the door. The code became the source of truth by default -- not because it's good at the job, but because everything else died first.
Amazon's fix was more review gates. That doesn't scale. The real lesson: AI agents need something to reconcile against before they touch production.
AI is making rewrites cheap. But 70% still fail. The code is disposable. The contracts are not.
Bill Eisenhauer proposed that specifications are the durable artifact and code is a regenerable rendering. The natural next question: what does the infrastructure look like that makes that work at enterprise scale?
Every compliance tool tells you you're broken. None of them fix it. What if a single declaration triggered automatic, codebase-wide reconciliation?
One agent is a tool. Many agents without shared identity is chaos. The multi-agent era needs a single source of truth about what the software is.
Specs drift because they describe states. Deltas survive because they describe decisions. Stop writing better documents. Start declaring better deltas.
When AI makes execution cheap, the scarce resource is not typing. It is deciding. The discipline of the next era is knowing the difference.
MCP has 97 million monthly SDK downloads and no built-in identity layer. That's a problem.
We use cookies to understand how you use ribo.dev and improve your experience.
Learn more in our Cookie Policy