Guide
AI for startup founders: start with your decisions
The most useful first job for a startup founder's AI is the decision you keep putting off. Paste your notes into ChatGPT or Claude and ask for a short decision memo: the recommendation and what you need first, only the context the reader needs, two to four real options (including doing nothing) with what each costs and gives up, the top risks with an early warning sign each, and who decides by when. It turns a week of going back and forth into a page a co-founder can answer in ten minutes. Say whether the choice is easy to reverse: Jeff Bezos's 2016 letter calls reversible decisions "two-way doors" that can be made quickly. Keep other people's personal data and confidential terms out.
Short answer
- Start with one stuck decision.
- Ask for a memo: recommendation and ask first.
- Two to four options, with what each gives up.
- A decider and a deadline.
The brief to copy
Decision memo brief
"Write a decision memo from my notes, under two pages.
Start with one paragraph: the decision, your recommendation, who decides, by when, and what it costs not to decide by then.
Then only the context the decider needs, with numbers and where they come from.
Give two to four real options, including doing nothing if that's possible: what it is, cost, time, what it gets us, what it gives up. Use a table for the numbers.
Say why the recommendation wins and what we give up by choosing it, and whether it's easy to reverse. Name the top two or three risks, each with a warning sign.
End with exactly what I need and the first step after a yes. Write in sentences; no adjectives instead of numbers.
My notes: [paste]"
A worked example
Illustration: a made-up business. The founder of a six-person software startup has been going back and forth for three weeks on whether to hire a second engineer. From her notes, the memo opens:
From the memo
Decision: hire a contractor for three months, not a full-time engineer yet. Decider: both co-founders, by Friday 13 November. Not deciding costs about two weeks of the release date for every month we wait.
Options:
A. Full-time engineer: about 3 months to hire; gives a long-term owner; costs the most and is hard to reverse.
B. Contractor for three months: starts in 2 weeks; ships the release; easy to end. Gives up: knowledge leaves when the contract ends.
C. Do nothing: no cost now; the release slips about a quarter.
Why B: the release date matters most this quarter, and B is a two-way door: if the contractor works out, we can offer a full-time role in month three.
Risk: code nobody else knows. Warning sign: no pull request reviewed by the team in the first two weeks.
Ask: approve the contractor budget; first step, post the brief on Monday.
Her co-founder replies with one question about the budget and a yes the same afternoon. She checks the release estimate in the memo against the engineer's own estimate before sending it.
Before you start
- Pick one decision that has waited more than a week.
- Paste your notes, not a polished summary; the memo does the sorting.
- Check every number against where it came from.
- Keep out other people's personal data and anything under confidentiality; see what not to paste.
- On a personal plan, turn off model training in settings first.
Next step: a project that knows your company
Pasting the same background into every chat gets old fast. A Claude project for founders sets up one place with your company's stage, metrics and rules, so every memo and update starts from them.
Skills for this
- ManagementDecision memoWrites a short narrative decision memo: recommendation first, context, options with trade-offs, risks, and the ask with a decider and deadline.
- FinanceMonthly investor updateWrites a short, honest monthly investor update: summary, KPIs, highlights, lowlights with cause and fix, cash and runway, and specific asks.
- Thinkers' principlesBezos-style one-way or two-way door checkSorts pending decisions into one-way and two-way doors, matches each to a careful or a quick process, and checks the information level, disagree and commit, and what needs escalating.
- FinanceCash runway updateWrites a cash runway update: cash, 3-month average net burn, runway in months, base, downside and upside scenarios, and levers, formulas shown.
Questions
How can startup founders use AI?
Start with the decisions you're stuck on. Paste your notes and ask for a short decision memo: the recommendation and what you need first, only the context needed, two to four real options with costs and trade-offs, the risks, and who decides by when. Writing it down is what makes a decision fast, and a co-founder or investor can answer it in minutes.
What should a founder not paste into ChatGPT?
Anything you couldn't show an investor or employee without permission: other people's personal data, customer lists, passwords, and term sheets under confidentiality. On a personal plan, turn off model training in settings first, and move company work to a business account once others join.
Can AI make the decision for me?
No. It organises the choice: the options, what each costs and gives up, the risks. You still decide. Say in the memo whether the decision is easy to reverse; Jeff Bezos's 2016 letter calls those "two-way doors" that can use a light-weight process.
AI skill
Have your AI write the memo that gets a decision made
Put this skill in AI Skill Pod and everyone on your team gets it in their Claude, Cursor or ChatGPT (Pro, Business, Enterprise): one invite link, free to start. Decision memo has your AI write a short decision memo from your notes: the recommendation and ask first, the context needed, two to four options with costs and trade-offs, the top risks with warning signs, and a decider and deadline.
14 days free · No card needed
Sources
- Amazon: 2016 letter to shareholders (Jeff Bezos) — "Many decisions are reversible, two-way doors. Those decisions can use a light-weight process"; most decisions with around 70% of the information you wish you had (read 2026-11-05)
- OpenAI Help: Data controls — on Free, Plus and Pro personal accounts, model training is on by default and can be turned off; read through search-result text (read 2026-11-05)
- Anthropic Privacy Center: Is my data used for model training? — personal plans: your choice in settings (read 2026-11-05)