Archive

Most enterprise AI pilots don't fail dramatically; they expire quietly. The reason isn't model quality or data — it's that nobody owns the seam between the AI capability and the organizational process it's supposed to serve.

What happens to designers, product owners, and engineers when AI sits at the center of the production process, and what scooting toward each other actually looks like in practice.

Most enterprises assumed closing the gap between what an AI knows and what's actually true right now was a software problem. It's turning out to be a data problem, and it's considerably more work than the initial project scope assumed.

AI-assisted development is making individual changes faster, but who's measuring whether the codebase is drifting structurally? This post names the missing instrument: a Structural Divergence Index for detecting pattern entropy, coupling drift, and boundary erosion before they become the rewrite.

The AI productivity promise is real. The AI civilization promise of curing Alzheimer's and cracking fusion is being made against a computational credit line that doesn't exist yet. Here's the engineering reality underneath the marketing.

A philosophical argument for the engineers who are quietly nervous, and the ones who should be.

Tekhton v1 could execute tasks. v2 can pursue milestones to completion, recover from failures, split oversized work, and harden itself against the security problems I didn't know I had. It also built most of itself.

A proposal for native AI capability discovery and intent as a first-class HTTP primitive, replacing MCP's runtime protocol with a static JSON manifest at /.well-known/ai-capabilities for HTTP-callable APIs.

The accessible version of the proposal: why AI agents need a simpler way to discover and use web APIs, explained without the jargon.

Building Tekhton, a multi-agent dev pipeline, forced me to learn agentic systems from the ground up. Here's what survived contact with real engineering constraints.