← All topics

Working with AI

Prompting, memory, and getting real work out of the model, including knowing when not to trust it.

Showing 1–20 of 98 posts

  1. Ask your AI to save its work as files

    I've kept a daily log in a file since the early days of using models, so I can switch models or start a new chat and pick up where I was. Ask for your work as files, and keep them in named folders.

    3 min read

  2. Plan big, do small

    A big model should plan the work and find the tools, libraries and free databases that already exist. Smaller models, or programs on your own computer, do what I like to call the donkey work.

    5 min read

  3. School family knowledge base, part 2

    Part 2 of a build log about a school family knowledge base: what information each of the eight sources holds, and the format it arrives in.

    4 min read

  4. School family knowledge base, part 1

    Schools send us information in eight places. That's a lot of channels for parents to coordinate and understand. Part 1 of a build log: the main issue, and how the tasks have to be broken down.

    2 min read

  5. The 3,848-Message Sieve

    What It Takes to Let an Agent Clean Your Inbox

    3 min read

  6. The Infinite Sandwich and the Made Bed

    Why Systems Thinking Is Your Real AI Superpower

    5 min read

  7. Hey, That's Not the Real Batmobile!

    Why "You Are A Marketing Expert" Isn’t A Prompt

    4 min read

  8. Gems Are Becoming Skills, but I Can Find My Gems, Not My Skills

    I am using Gemini more and more for editorial and creative work. I like it for developing ideas and roughing out illustrations, while I tend to use other tools for purely operational jobs. That has meant spending quite a bit of time playing with Gems.

    2 min read

  9. Space to be human

    A week of new school routines and a storm and flood, and what I needed was space to be human.

    3 min read

  10. How developers organise and maintain agent skills

    An Ask HN discussion describes skills as repeatable workflows, with shared repositories, human review, behavioural tests and updates across coding tools.

    4 min read

  11. Handing off the 60%

    Planning is 40% of a task; execution is the 60% that eats your week. Plus vibe sourcing — finding the open-source tool instead of generating one.

    7 min read

  12. SoN 2.31: Don't be a meat proxy

    A meat proxy forwards an AI's answer without reading it. That hands the next person more work than the question did.

    10 min read

  13. Five words for the same thing

    The jargon is doing less work than it looks.

    8 min read

  14. SoN 2.30: Which of your AI's rules are still doing a job?

    Last week's advice and Anthropic's both hold. The test is one question you can ask about any rule you've written.

    12 min read

  15. From Custom Code to Conversational Prompts

    Who's allowed to build software has changed hands three or four times in a decade. Grok Build is the latest handoff.

    7 min read

  16. The Donkey Work Doesn't Need a Genius

    A $500 fine-tune of a 9B open model beat every frontier model on a catalogue-review task. I haven't trained anything.

    4 min read

  17. Rehearse the Deck You Didn't Write

    The illusion of explanatory depth is what happens when the first test of your understanding is live, in front of people.

    3 min read

  18. Kimi K3 Is Open. I Still Can't Run It.

    2.8 trillion parameters, 1.4 terabytes of memory to run it. Open weights existing and open weights being usable have stopped being the same claim.

    4 min read

  19. "AI Found a Security Bug" Is the Wrong Headline

    Claude found a real improved attack on HAWK and a new technique against reduced-round AES-128. The interesting part isn't that it found them.

    4 min read

  20. Hubble, and the Assumption Baked Into a Notes App Now

    "The best notepad for you and your agents." Hubble designs the note format around a second reader from the start.

    3 min read