- 5 min read
When AI Agents Act Without Permission
Autonomous AI agents can plan and execute tasks independently, but building trust and safety into their decisions is harder than giving them freedom.
Read the post → - 5 min read
What to Say When a Client Expands Scope
Scope creep kills timelines. Refusing it damages relationships. Show clients the cost in their terms, then let them decide what matters most.
Read the post → - 5 min read
Signal a deadline slip early, before certainty
Tell your client a deadline is at risk three weeks out, not three days. Early signals let them treat it as planning. Late ones feel like hidden failures.
Read the post → - 5 min read
When to Use Claude Cowork Instead of Chat
Chat is for thinking. Cowork is for producing. What changed when I moved architecture planning and project documentation out of a chat window.
Read the post → - 5 min read
How to Disagree With a Client Who Did Their Homework
Four objections to a client's own technical plan, all correct, all accepted. What made them land was not the analysis but the shape of the disagreement.
Read the post →
Blog archive
Writing on Delivery
A post-incident review is not an account of what happened; it's a change to how the system or the process works, and if it doesn't produce one, the review failed. Restore first, understand second: the sequence matters more than the narrative you build afterward. The difference between an incident that makes the next one less likely and one that just makes for a good story is whether it ends in a deployment checklist or a slide.
5 posts
5 posts