I write about shipping software and AI systems without hand-waving privacy or business risk.
Topics: architecture trade-offs, agents in production, data/privacy in AI workflows, and B2B product lessons.
If that’s useful, follow along — questions and pushback welcome.
Rebuilt https://t.co/SJTIqRx9ny.
AI that runs your operations, not just your demos. Process automation first, then web and mobile apps.
Led by a principal engineer. Shipped to production.
https://t.co/pyzVP9ukXQ
The moment an agent can open a PR, you’ve left “chat” and entered change management.
Who reviews? What can it touch? How do you roll back a bad tool call?
Prompt quality matters. Control plane matters more.
Shipping agents isn’t prompt engineering. It’s product + ops: tool permissions, idempotent side effects, audit logs, and a kill switch when the loop goes wrong.
Most “AI features” fail because the architecture was bolted on. Treat the model as an unreliable dependency: timeouts, evals, and a human path when confidence is low.
CoWe recently implemented AI agents for a 10M+ lines codebase. Here are some insights we had from the experience, drawbacks, and productivity gains: A Practical Look at Implementing AI in Software Engineering https://t.co/lE6zy2a115