Free guide · Thinking methods for work
Thinking-Method Menu
Choose a useful way to examine a work problem, then give the AI the facts, constraints and evidence it needs. These labels are thinking lenses—not hidden AI modes, expert endorsements or substitutes for judgment.
1. Start with the job, not the label
Write the decision or deliverable in plain language. Add the facts you know, the constraints you must respect and the assumptions you want challenged. A thinking label changes the questions the AI asks; it does not make missing evidence appear.
WORK PROBLEM I need to: [make a decision, prepare a proposal, solve a process issue or draft a plan] The result will be used by: [audience] The decision or next step is: [specific outcome] KNOWN FACTS - [fact — source] - [fact — source] CONSTRAINTS - [budget, deadline, policy, capacity or approval boundary] ASSUMPTIONS TO TEST - [assumption] Use the [METHOD] thinking lens below. Separate: 1. facts supported by my material, 2. inferences you are making, 3. missing information or evidence, 4. questions I need to answer. Do not invent figures, customer results, quotations, deadlines or approvals.
2. Pick one of the ten methods
| Method | Use it when you need to | Add this instruction |
|---|---|---|
| First principles | Separate the problem from habit and inherited assumptions | Break this into facts, constraints and assumptions. Identify the smallest parts that must be true before a solution can work. |
| Y Combinator | Pressure-test demand, experiments and growth logic | Identify the user problem, current evidence of demand, the smallest useful test and what result would change the decision. |
| Gary V | Consider attention and audience building before choosing what fits work | Identify who needs to notice this, why they would care, which useful message earns attention and which channels fit our workplace. Do not imitate a person’s voice. |
| ELI10 | Make an unfamiliar idea understandable | Explain this simply with one familiar example, then list the important nuance the simple explanation leaves out. |
| Red team | Find weaknesses and strong objections | Challenge the proposal. Identify failure modes, opposing evidence, affected stakeholders and the strongest reasons not to proceed. |
| Lindy | Examine practices that have lasted | Identify which parts rely on durable methods, what has changed and whether the older approach still fits this situation. |
| 80/20 | Focus effort on the few actions most likely to matter | Rank the small number of actions likely to create most of the useful result. State the evidence and uncertainty behind the ranking. |
| Human | Make a draft sound natural and usable | Revise against my writing criteria. Preserve the meaning, request, dates and evidence. Flag anything you would need to invent. |
| Anti-advice | Test when the standard recommendation fails | State the usual advice, the conditions that make it useful, the conditions where it fails and a safer alternative for those cases. |
| Unlearn | Revisit assumptions that may no longer fit | List the assumptions embedded in our current approach. For each, show why it existed, what has changed and what evidence would justify replacing it. |
Use a name as shorthand for the lens, not as a claim that the person or organization endorses the answer. Ask for the reasoning you need rather than “act as” a public figure.
3. Worked example: improve a proposal
This is a fictional internal example. It demonstrates how three methods reveal different issues without changing the underlying facts.
SOURCE A — Fictional intake-workflow proposal Problem: Internal service requests arrive through email, chat and a shared form. Observed sample: 40 requests reviewed from the last four weeks. In that sample, 14 requests were returned because required information was missing. Proposed change: Use one revised form for a four-week pilot with one department. Capacity: One operations coordinator can support the pilot for two hours per week. Decision requested: Approve preparation of the pilot plan. Unknown: Total request volume, baseline completion time, adoption rate and privacy review. No time savings, cost savings or service improvement has been demonstrated.
Apply first principles and the problem becomes: what information is essential to start a request, who supplies it, and which constraints make one intake route workable? The 40-request sample is evidence of a missing-information problem, but not proof that a new form will solve it.
Apply red team and the main objections become: the sample may not represent normal volume; staff may keep using email and chat; privacy review is unresolved; and two support hours may be too little. These are questions for the pilot plan, not reasons to invent a positive forecast.
Apply 80/20 and the likely first actions are: identify the few missing fields behind most returns, test them with one department, and measure return rate and completion time. The ranking is provisional because the source does not break the 14 returned requests down by missing field.
The methods produce different questions while preserving the same evidence boundary.
4. Choose the lens for your proposal
| If your problem is… | Start with… | Then consider… |
|---|---|---|
| Vague or built on inherited assumptions | First principles | Unlearn |
| A new service, offer or internal initiative | Y Combinator | Red team |
| Hard for the audience to understand | ELI10 | Human |
| Too broad for available time | 80/20 | Red team |
| Based on a popular “best practice” | Anti-advice | Lindy |
| Technically sound but unlikely to earn attention | Gary V | Human |
| Ready for approval and needs a challenge | Red team | 80/20 |
Use one lens first. Compare a second lens only when it adds a genuinely different question. Ten labels in one request usually create a longer answer, not a better decision.
5. Run the evidence check
Paste the draft and its source material after this prompt.
Review this draft for evidence quality. Create a table with: - claim or recommendation, - source supplied, - whether the source directly supports it, - what is inference rather than fact, - missing evidence, - correction or safer wording. Check figures, dates, customer or employee outcomes, savings, deadlines, quotations, approvals and words such as proven, guaranteed, always, never and will. Do not search for evidence unless I ask. Do not treat repetition across AI answers as verification. If the source is missing, label the claim unsupported rather than completing it from general knowledge. End with: 1. claims safe to retain, 2. claims that need qualification, 3. claims to remove, 4. questions the decision-maker still needs answered.
For the fictional proposal, “14 of 40 sampled requests were returned for missing information” is supported. “The new form will reduce delays by 35%” must be removed because no result or forecast supports it.
6. Review before you use the answer
- The job, audience and decision are explicit.
- Facts, constraints and assumptions are separated.
- The chosen method fits the problem.
- The answer shows its reasoning instead of relying on the label.
- Important claims point to supplied evidence.
- Missing information remains visible.
- Recommendations do not become approvals.
- Figures, dates and commitments have not changed.
- A second method adds a different perspective rather than more words.
- A person accountable for the work makes the final decision.