The Blog

Practical AI, in writing.

Workflows, prompts, tools, and field notes from building production AI systems.Page 11 of 19

Correct ChatGPT once and it apologizes forever. One line stops the groveling and gets you back to work.

You set instructions and ChatGPT ignores them anyway. The defaults are quietly winning. Here is how to make your rules stick.

Long ChatGPT conversations slowly wander off track. One sentence pulls it back and keeps it on the original goal.

OpenClaw gets way more powerful with the right apps wired in. Here are five free-tier tools that turn it into a real builder.

ChatGPT gives you the answer but not the why. Three lines make it show its reasoning so you can trust the result.

You do not need to master AI before you use it. Waiting to feel expert enough is the thing actually holding you back.

Reports say the Pentagon pushed Anthropic to loosen its AI safety limits. Anthropic said no. Here are the two red lines it would not cross.

Act as an expert is the most overused prompt opener and it barely does anything. Here is what to say instead for better output.

Bad ChatGPT output usually means a bad prompt. The first three lines control everything. Here is the structure that fixes it.

If ChatGPT suddenly feels off, it's probably pulling from stale saved Memory. Here's the 30-second cleanup that fixes it.

ChatGPT will happily make up a statistic and cite a source that never existed. One line in Custom Instructions shuts that down for good.

Most people jump straight into prompts and wonder why the answers are generic. Spend 90 seconds setting up a Project and every chat starts smart.