The Recursive AI Method

The Recursive Method, a widening three-arc spiral: inner arc Output recursion (within a session, three spokes — missing? sceptic? second-best?); middle arc Process recursion (session-to-session — the question behind my question); outer arc Practice recursion (week-scale — read back the week). Centre: compounding thinking. Outer-edge dashed line: single-pass plateau.

​

Output recursion. Feed AI’s output back to AI with a sharper question. Three prompts work for almost every output: what is this missing?, what would a sceptic say?, what is the second-best version of this that I would not have thought of? Each operates on a different weakness in the first pass, what is absent, what is contestable, what is unimagined. Run all three and the leader ends up with a different answer altogether, well past a polish of the first one.

​

Process recursion. After output recursion, turn AI on the leader’s own prompting. Looking at how I asked the original question, what was I assuming that I should have surfaced? What was the question behind my question? The next prompt, written after process recursion, is a different prompt. Over a few weeks, the leader’s prompting gets sharper.

​

Practice recursion. At the end of the week, ask AI to read back the week’s conversation log and identify patterns the leader would not see from inside it. What did I outsource that I should have done myself? What kind of question came up repeatedly? What is one move I made that, if made routinely, would change the quality of my AI use? Schön’s reflection-on-action (1983) applied to AI use itself.

The three compound: output recursion makes any single session deeper; process recursion makes the next session start sharper; practice recursion makes the next week start more deliberate.

​

Genesis

Stacks George Polya’s How to Solve It (1945), the looking-back fourth phase that most problem-solvers skip but that produces mathematical maturity, onto Donald Schön’s The Reflective Practitioner (1983) reflection-in-action and reflection-on-action distinction. Polya (1945) gives the move at the level of the single problem; Schön (1983) at the level of the practice.

​

Why it matters now

Single-pass AI use plateaus quickly because nothing it produces sharpens what comes next. The leader’s prompts in week twelve look like the prompts in week one, the outputs are equivalently competent, and the judgement about when to trust AI has not grown because it has not been tested. The Recursive Method is the small structural difference that determines whether the second quarter looks like the first, or like a different practice altogether.

​

The moves

This week: one second-pass on every substantive AI session before using the output (default prompt: what is the strongest objection to this answer that I have not addressed?). One process-recursion question after each substantive session, asking AI how the original prompt could have been better. A Friday practice-recursion review, fifteen minutes, asking AI to read back the week’s conversation log and surface patterns.

​

How ChangeSchool applies it with executives

We run senior cohorts through the Three Recursions Audit in cohort sessions: each leader brings one substantive AI session from the previous week, runs all three recursions on it alongside a peer, and compares the post-recursion artefact against the pre-recursion one. The before-and-after comparison is the experiential move; abstract argument for recursion never lands as hard as seeing what the second-pass prompt does to one’s own draft.

​

The discipline

A second-pass default, over weeks, the second-pass prompt becomes the leader’s reflex; the first AI answer feels incomplete in the way a first draft feels incomplete. A prompt journal, a few lines a week from the process-recursion answers, surfacing the patterns in the leader’s own prompting that would otherwise repeat indefinitely.

​

A Friday cadence, the fifteen-minute practice-recursion at a fixed end-of-week moment, before logging off. Without the cadence, the recursion does not happen; with it, twelve sets of patterns are surfaced over a quarter and twelve adjustments made to the following week.

​

​

‘The first AI answer is the start of the work, not the end of it.’ 

​

​

​

​

​

​

Viren Lall, Managing Director,

ChangeSchool LDN (2026).

virenlall.com/recursive-ai-method

​

​

​