Overview
A recurring, cross-domain pattern emerged from a 2026 HN discussion of cognitive offloading for how to use AI to actually learn something rather than merely produce an output that looks learned: get the information from the AI, then put it into practice yourself. Commenters converged on this rule independently across programming, gardening, cocktail-making, and general skill acquisition.
The practice-after pattern
- One commenter described contractors renovating an old house who hit an unusual problem outside their experience. They consulted ChatGPT for a solution and code-compliant materials, verified the claims against material specs, and only then did the work themselves. “This is the right way to benefit from LLMs in learning something: Get the information then put it into practice so you learn it… It’s like skimming a math textbook but then skipping the exercises and never doing any quizzes.”
- Another commenter described learning gardening the same way: “I definitely retain it. In the case of the gardening — I pretty much immediately go outside and practice it… how I get the info doesn’t matter (book, video, llm) because I’m locking in the skill by doing the thing in the garden.”
- A third commenter used an LLM to improve an old fashioned cocktail recipe, verified the result by taste (“skimping on the orange oil/zest makes a huge difference”), and considered the knowledge locked in after one practical test — a low-stakes case where the learning loop closes almost immediately.
A framework for using AI without ceding thinking
One commenter laid out a discipline for using AI to “extend my capabilities” rather than “remove my agency”:
- Debug — let the AI hunt for things you’d miss.
- Review — ask it to critique finished work for an outside perspective.
- Implement an already-decided design — “I have already gone through the thoughtful engineering to decide what to do and how to do it. The rest amounts to translating pseudocode… Let it type what I would have typed anyway.”
- Suggest ideas — treat suggestions as one more input, not the verdict: “I don’t take the AI at its word and let it decide what is best for me.”
Their summary maxim: “I am not replacing learning, thinking, or deciding.”
A related, more compressed maxim from another commenter: “If you offload any of your thinking to AI, you’re offloading too much of your thinking to AI. Offload your execution, not your thinking.”
Resources
- 2026-07-15 ◦ Are we offloading too much of our thinking to AI? (HN discussion) — source of the practice-after pattern (contractor, gardening, cocktail anecdotes) and the debug/review/implement/suggest framework and offload-execution maxim above
- 2026-07-15 ◦ Cognitive offloading — the hub topic covering the broader debate this practical framework responds to
- 2026-07-15 ◦ AI coding assistants — see “Agency-preserving usage patterns” for the programming-specific application of these ideas