Strategy
Keep who your product is for, and what it is not.
Product strategy is a handful of positions you hold for a while and then change — who the product is for, what it will never do, what you charge and why. You work most of them out in ChatGPT or Claude, an evening at a time, over months. Then a customer says something in June that turns a position you settled in the spring into the wrong one. You change your mind. And the position you’re on now exists nowhere but your head.
Strategy decisions don’t stay decided
Most decisions hold still once you make them. These don’t. You settle who the product is for on the evidence you had at the time — four interviews and a hunch — and a year later you have forty more interviews, nine paying customers and a churn reason nobody predicted. The position that was right then is wrong now, and changing it is the correct thing to do.
So you change it in a ChatGPT or Claude session, the way you worked it out the first time. Fourteen months in, that’s four positions reversed, ninety-odd sessions and no page anywhere that says which version you’re on. And the expensive loss isn’t the new answer. It’s the old one. Six weeks later you can’t remember whether you dropped the multi-store segment because of the sales cycle or because of the support load, so you can’t tell whether the reason still holds — and you drift back toward it one feature at a time.
Where the version you’re on lives
Giraffe holds the positions themselves, one live value each, next to the calls you’ve settled and the questions you haven’t. When one moves, you tell your assistant: “Update the positioning — single-location shops only, not chains. Save why.” The new wording takes the row. The wording it replaced is kept underneath, dated, with whether it was you or an assistant who wrote it.
Nothing is watching for a contradiction, and Giraffe will never ask you to confirm a value. What it gives you is both versions on one page — so the next time you catch yourself arguing for the old thing, you can go and read why you dropped it instead of arguing it out again from memory. Most weeks that’s the whole job, with no AI session at all. Here is how Giraffe works, step by step.
Fourteen months of a product, in one project
Say you make repair-desk software for bike shops. You’ve been at it since last spring, mostly in evenings — one working out whether the multi-store chains were worth chasing, a long one after two shops left in the same week, another on whether $29 a month covered the support you were doing yourself. Dozens of sessions, two assistants, no document. Here is what all of it landed on:
Spoke — who it’s for
- Current
- Decisions
- Parts belong on a job after all. “No inventory, ever” was right about stock counts and wrong about parts — eleven of the fourteen spring trials asked where a $52 cassette goes on the bill, and two of the shops that left named it on the way out. Reversed in June, and only that far.
- One plan, not three tiers. A shop with two mechanics and a shop with four have the same problem, and tiers made people answer “which one am I” before they had seen it work.
- No free tier. Every shop still paying came off the fourteen-day trial; the free accounts opened in January are still open and still empty.
- Open questions
- Whether “not for chains” holds the first time one offers to pay. Two have asked this year. It is an easy line to write down and a hard one to keep in a month where nine shops are paying.
- Whether $39 is underpriced for a shop billing $8,000 a month in labour, or about right for the one billing $2,000.
- Notes
- All nine paying shops arrived the same way — one mechanic telling another in a Facebook group. Nine for nine. No idea yet whether that is a channel or just the first nine.
- Recent changes
The reversal is the part worth keeping
The middle of that card is a decision that got un-made. “No inventory, ever” was a good call in March: it kept the product to one screen and one job, and it kept you out of a fight with the point-of-sale system every shop already owns. By June, eleven of the fourteen trials that spring had asked the same question about where a $52 cassette goes on the bill, and two of the shops that left named it on the way out. So it reversed — narrowly. Parts attach to a job. Nothing is counted or reordered.
Both halves are on the page. The Scope row says what is true now. The history says it used to read “jobs and status only, no parts,” and when that stopped being true. The next time a shop asks for stock counts, you aren’t working from a vague sense of having decided something once: March’s reason and June’s evidence are both in front of you, and you can see that the second one overturned half of the first and left the other half standing. A position you have reversed once is the one you are most likely to reverse back by accident.
Price moved last, after the customer did
Price is one row of four on that card, and it was the last thing to change. $29 went to $39 only last month — after who the product was for had narrowed in March, not before. A one-shop owner billing $2,000 a month in labour is a different buyer from a four-store chain with a purchasing process, and a flat plan with no per-mechanic seat is what the first one will agree to on a phone call. When the pricing question comes back, and it comes back every quarter, the number isn’t the useful part of the record. The row above it is. A reversal like this one is written up as it happened, week by week, in a positioning call that took five weeks.
A few more questions
- Isn’t this what a strategy doc is for?
- A strategy doc tells you what you think today. It doesn’t tell you what you thought six months ago or what changed your mind, because the old version was overwritten to keep the doc readable. A Giraffe project holds one live value for each position — who the product is for, what it won’t do, what you charge — and keeps the wording each one replaced, with the date it was replaced on.
- What happens when I reverse a decision I already saved?
- You tell ChatGPT or Claude the new position and it takes the row. The one it replaced is kept in the project’s history, dated, with whether you or an assistant wrote it. That means a reversal reads as a reversal — you can see the call you made, the call you make now, and how long the first one stood — rather than as a document that always agreed with you.
- Can my co-founder read the project?
- No. A Giraffe project belongs to one person: there are no shared projects, no team accounts and no administrator. If you want someone else to see one piece of it, a single document can be shared read-only by link. Everything else stays yours, and you can delete any of it whenever you want.
- Will it tell me my new position contradicts an old one?
- No. Giraffe doesn’t watch your projects, doesn’t judge what you save, and will never ask you to confirm a value. It records what you told it, in order, with dates. Noticing that June contradicts March is your job — or the job of whichever assistant you ask to read the project — and having both versions on one page is what makes it possible.
A reversed decision is still a decision.