Thoughts on software, business, and building things that matter
AI coding agents are becoming part of the delivery workflow. The teams that benefit most will not be the ones that trust them blindly, but the ones that put policy, review and operational controls around them.
AI tools can get a prototype running quickly. The serious work starts when customers, data, operations and future changes depend on it.
Pwn2Own Berlin is testing Codex, Claude Code and Cursor. For teams adopting coding agents, the lesson is clear: treat the agent runtime as privileged software, not just a smarter editor.
The current AI Act discussion on X is a reminder that companies should not treat AI compliance as a late legal cleanup. It starts with product architecture, data flows and responsibility boundaries.
AI app builders make internal tools and prototypes visible very quickly. The real risk appears when nobody knows which apps exist, which data they touch and who owns them.
AI can help find security bugs faster. The business value comes from harnesses, reproducible tests, triage and release processes around the model.
AI agents can be useful in real business workflows, but only when access, auditability, approval boundaries and accountability are designed before production rollout.
The current spec-driven development discussion is a useful reminder: when AI makes implementation faster, unclear product intent becomes the expensive part.
Production-ready AI agents need more than a strong model. They need architecture around context, tools, permissions, validation and operations.