A lot of people try an AI model, get a disappointing result, and decide that it simply can’t do what they need.
That may have been true on the day they tried it. It may not be true now.
The models are improving quickly, the tools around them are changing, and we’re also getting better at explaining what we need and managing longer AI-assisted projects.
This week’s guide, A Failed AI Experiment Isn’t a Final Answer, looks at how to revisit an unsuccessful AI task and determine whether the model, the method, or both have improved.
https://askgoodquestions.dev/guides/a-failed-ai-experiment-isnt-a-final-answer
The companion field note, Try, Try Again, includes some of what I’ve been seeing firsthand while working with Claude and ChatGPT. That includes a recent programming session where ChatGPT created a 10,000-row test case, discovered a timeout problem, and revised the search design around the results.
https://askgoodquestions.dev/field-notes/try-try-again
With AI, failure has an expiration date. The model that disappointed you a few months ago may no longer exist in any meaningful sense.
Try the old task again. You may be surprised by what happens.


