Product retrospectives and methods for working with AI. Completed experiments and proposed designs are labeled separately.
New here? Start with the reading guideWould an AI solution to a Millennium Prize Problem let us compute the entire world? Even if it did, why should that mean we live in a simulation? I want to work through the reasoning behind Musk's claim.
Writing an event description, finding this week's schedule and checking availability all happen in one chat, but need different capabilities. Follow those jobs to see how language models, RAG, agents, MCP and Skills work together.
Two coffee shops can quickly produce beautiful promotional images. Why would a customer choose one? As production capabilities converge, a brand's promises, the product's ability to keep them and what makes customers hesitate deserve more attention.
A clip can look great and still make little sense as part of a story. Change an earlier shot, and later ones need checking again. A director Agent's usefulness depends on how it handles those revisions—and whether experience from one draft helps with the next piece.
An employee moves temporarily to another city, and different payroll rules apply. A program can handle the calculation. Someone still has to confirm the rules, update them, and check the changes after handover.
Change one narration passage, and the sound, subtitles and later scenes may all need updating. Where does the model need to decide again, and where can software carry on? Whether fewer tokens save money depends on the rest of the edit, too.
The conversation is saved and the steps are recorded. Change the source material, and someone still has to explain the job again. Recording the conditions and decisions behind a good result helps us see how much of it we can repeat.
The same meeting notes can help someone catch up or help a project lead act. If that purpose stays unstated, even a thorough AI summary may be of little use.
Three drafts arrive in minutes, but choosing one still takes time: read the differences, check the facts and explain the choice to colleagues. Why can AI save the writing while leaving us just as busy?
