The 1% Rule for AI

The One Percent Rule compounding curve over 156 weeks. Line A: 1.01 per week, embedded (rises to ~1.67 at week 52, ~2.7 at week 104, ~4.5 at week 156). Line B: read-about-but-not-used (flat at week-one capability). The gap between the curves is what skipping embedding costs.

​

Embedded before the next. The improvement must be used at least three or four times in real work before another is added. Improvements that are read about and not used decay inside the week. Improvements that are used three times in real work become part of how the leader operates by the following Monday.

​

One per week. Not three, not five. One. The arithmetic, popularised by James Clear (Atomic Habits, 2018): 1.01 to the power of 52 (working weeks in a year) is roughly 1.67. A leader who improves at one specific dimension each week, and embeds it, is around 67% better at AI by year-end than the leader who started in the same place. Held over two years, around 2.7×; over three years, around 4.5×. Numbers are illustrative; the mechanism, embedded gains compound, un-embedded gains do not, is robust.

​

Genesis

Two prior bodies of work sit underneath. Sir Dave Brailsford’s marginal-gains doctrine (Team Sky, 2010) showed that the aggregation of small improvements, each beneath notice individually, produces multiple Tour wins in aggregate. Anders Ericsson’s deliberate-practice literature (1993) separates ordinary repetition from work on a specific identified weakness with feedback before moving on. James Clear (2018) gave the compounding arithmetic its popular form in Atomic Habits. The 1% Rule for AI is what marginal gains, deliberate practice and Clear’s 1.01^52 logic look like applied to a leader’s personal AI use.

​

Why it matters now

AI use is invisible by default. Without a tracked log, the leader has no way to tell whether anything has changed week-on-week. The wholesale-better-at-AI ambition, a course, a Saturday strategy session, requires resources the leadership calendar does not provide. The result, six months in: a thousand AI sessions, one week of practice repeated a hundred times. The 1% Rule replaces the ambition with a unit of progress that fits the calendar.

​

The rule starts with the leader interrogating AI for usable patterns; over time the relationship inverts and the leader begins to teach AI the standards, vocabulary and judgement of their own work. The gains compound when the second mode kicks in.

​

The moves

Pick the dimension, ten minutes on Monday, one specific nameable improvement. Use it three times during the week, on real work whose output matters. Name the improvement on Friday in four lines: what it was, where it worked, where it did not, what next week’s will be.

​

End each substantive AI working session with three closing prompts that build the second mode: What did you learn from our interaction? What can I learn from our interaction? What can you commit to your memory so that we can benefit from our collective learning? Three lines a week from these answers become the core of the prompt journal.

​

The four-line Friday note is not a record. It is a decision input. At the next Monday’s ten minutes, the leader reads last Friday’s note before picking the new dimension, and any open did not work line carries forward until it resolves. The note is what makes the compounding visible.

​

How ChangeSchool applies it with executives

We run senior cohorts through the Marginal Gains Audit: each leader keeps the four-line Friday note for the duration of the cohort, brings the log to the next cohort session, and reads it back to a peer. The rule holds because the log is read aloud.

​

The discipline

A weekly micro-improvement log, five lines a week, re-read once a month, captures the improvement and makes the compounding visible. A before-the-week practice question on Monday: what is the one thing I want to be doing differently with AI by Friday? A named-improvement test on Friday, the leader has to name, in a complete sentence, one thing they are doing differently with AI than they were the previous Friday. Three weeks of failing the test means the rule has stalled and needs the dimension reset.

​

“One named improvement a week, embedded before the next; the smallest unit of progress that will actually be made, held long enough to compound.”

 

 

Viren Lall, Managing Director,

ChangeSchool LDN (2026).

virenlall.com/one-percent-rule-ai

​

​

​