A candid look at our real AI workflow
Video coming soon
No YouTube video has been added yet.
No theory this time. Paul and Daniel open up their actual day-to-day: where Claude saves them hours, where they deliberately keep it out, and the habits that keep the work high quality.
Key ideas
Show notes
Everyone talks about AI in the abstract. This episode is concrete: Paul and Daniel walk through their real workflow, the specific tasks they delegate to Claude, the ones they never would, and the guardrails that keep the output trustworthy.
Timestamps
Quotable
“Cold prompts get cold results, every time.”
— Paul Maxwell
Full transcript
Paul: People assume we use Claude for everything. We don't. We use it for a specific band of work: drafting, exploring, restructuring, summarizing. The stuff where a strong first pass compounds.
Daniel: And there's a whole category we keep fully human. Anything where being confidently wrong is expensive, or where the judgment is the actual product.
Paul: The consistency comes from how we start a session. Context first, clear task, then iterate. Cold prompts get cold results, every time.
Daniel: And we assume the model can be confidently wrong. So we read everything. The speed is real, but it never removes the responsibility to check.
Related links