Chapter 10
Keep it: save what worked and finish your checklist
The chapter in a minute
- Save each prompt that worked with the same five fields: task, prompt word for word, what worked, what to change and date.
- What you save feeds the next Know it, so each time a job comes round it starts from what you learned last time.
- Give a repeating job a set day and start from the saved prompt in a fresh chat each time. Change one thing at a time so you can tell what helped.
- Write the minutes down, checking included. A feeling of speed is not the same as time saved.
- Checklist question 5 is 'Have I saved what worked in my prompt notebook, with a note on what to change next time?' Your card now has all five questions.
Try it now
Run the whole loop once
0 of 9 steps done
Step 1 of 9
Choose one small real task, such as a note to a neighbour, a shopping list for a party or a short notice for a club. If a notebook entry already fits the job, start from it.
Try this
Try this, from the book
The next time you use a saved prompt, note the time when you start and when you finish, checking included. Write the minutes beside that notebook entry. It takes seconds, so give it a go on your next task!
Tools and settings, as of a date
The book points to the ‘Checklist card and notebook’ section of the companion page. Here it is in full, with the date each detail was checked. Open it on the companion page
Chapter 10 points here: the printable card and the notebook page in the companion PDF. The card text itself is in the "Printable checklist card" section below.
The prompt notebook page
The five fields are always the same: Task, Prompt (word for word), What worked, What to change, Date. The context the prompt needed sits inside the Prompt line, so there is no sixth field. Write "name" and "date" in place of names and dates you won't reuse.
Keep it wherever you will actually open it: a paper notebook, a notes app or the notebook page in the companion PDF.
Download: the companion PDF
Optional: a study about feeling faster
Chapter 10 mentions one 2025 study of experienced software developers who felt faster with the tools while the clock showed their tasks took longer. Becker and colleagues, "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity", arXiv 2507.09089 (opens in a new tab), first posted 12 July 2025, revised 25 July 2025. 16 experienced open source developers, 246 tasks, early 2025 tools. They forecast a 24% time cut, estimated a 20% cut afterwards, and measured a 19% increase.
Checked: 2 November 2026. It is still a preprint: arXiv lists no journal publication. The same research group posted an update on 24 February 2026 about a larger follow-up begun in August 2025 (57 developers, more than 800 tasks). That round pointed the other way, with an estimated 18% speed-up for returning developers and 4% for new ones, but the group itself says those figures are unreliable, because many developers would no longer agree to work without the tools. It is now changing how it runs the study. METR, "We are Changing our Developer Productivity Experiment Design" (opens in a new tab), 24 February 2026. Read together, the two rounds back the book's point: how fast you feel is not a measurement.
Present the results together with the further reading, and with the line that results depend on the task, the worker and the tool, never one alone:
- Noy and Zhang, "Experimental evidence on the productivity effects of generative artificial intelligence", Science, July 2023: set writing tasks, 40% less time. doi.org/10.1126/science.adh2586 (opens in a new tab)
- Brynjolfsson, Li and Raymond, "Generative AI at Work", Quarterly Journal of Economics 2025: customer support agents resolved 15% more issues per hour on average. doi.org/10.1093/qje/qjae044 (opens in a new tab)
- Humlum and Vestergaard, "Still Waters, Rapid Currents", NBER Working Paper 33777, May 2025, revised March 2026: no measurable change in Danish earnings or hours. nber.org/papers/w33777 (opens in a new tab)
From the book
Your card has one question for each step of the loop. Each one takes a few seconds to answer, and a no tells you which step to go back to.
Sources
- Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity (opens in a new tab) · Becker, Joel and Rush, Nate and Barnes, Elizabeth and Rein, David, 2025