Thoughts on technology leadership, AI strategy, and building engineering organizations.
Code review, quality gates, and accountability norms were designed around assumptions about human authorship. Those assumptions don't survive contact with an agent running in a loop, and the failure looks like throughput going up while the gate quietly becomes a rubber stamp.
The AI Kill Switch Act regulates the model. The dangerous surface in real deployments is the operating envelope, and the operating envelope belongs to the deployer, whom the bill exempts by design.
The layer between your application and the model is where cost, latency, quality, compliance, and vendor sovereignty are actually decided. In most organizations, no one at the executive table owns it.
Per-seat AI pricing was never flat. The meter was just invisible. As of June 1, it isn't.
AI coding agents are improving every velocity metric while the underlying system becomes harder to reason about. Why throughput and comprehension are decoupling — and what to do about it.
The 95% of enterprise AI pilots that show no P&L impact may say less about the technology than about the SaaS-era frameworks we're using to measure it.
When an agent with persistent memory learns from a poisoned repo, reverting the commit doesn't revert the lesson. A new class of supply-chain risk hiding downstream of remediation.
AI has made knowledge workers faster, but the real value requires auditing the work itself — not bolting assistance onto workflows designed for humans doing every step.
Organizations evaluate AI tools by comparing models, but the interface layer — the client — is where adoption succeeds or fails. Why procurement needs to invert its process.
AI has broken the assumptions that made ARR meaningful. Pilots counted as recurring revenue, usage that swings 10x, and compute costs eating margins — and a framework for measuring honestly.
70% of AI implementation challenges stem from people and process issues. Only 10% involve the algorithms. Yet organizations spend 80% of their time on that bottom 10%.
Why most enterprise AI initiatives stall — and the organizational design pattern that keeps them moving.
A pre-deployment evaluation framework for agentic AI in the enterprise.