I want Pineapple to take on more of the work of running a business. The obvious place to test that idea is Pineapple itself.
Conceptual illustration. It does not show the Pineapple interface or a measured result.
We already have part of the system. Peter Assistant holds our plans, customer research, metrics, and recurring reviews. AI helps investigate problems and prepare work. But those pieces have mostly lived inside a private operating repository.
That makes it difficult for someone outside the company to understand what AI does, why we choose one task over another, or whether the work is helping.
So I am putting the approach into a public handbook.
What I want people to see
Who Pineapple is built for. How we decide what matters. Which numbers we trust and which still need checking. What AI can do independently, where I remain responsible, and what we learn when something fails.
I also want the guide to be useful beyond Pineapple. A service-business owner should be able to take one workflow, use their own sources, and try it without building a complicated system first.
PostHog's handbook is useful because it exposes the operating choices behind the product. Buffer's Open hub puts company results where anyone can find them. We are borrowing the habit of explaining the work and its evidence.
One task we can show
On 18 October, we checked our pricing FAQ against the domain guide. The FAQ could be read as promising a purchased domain. Our support facts distinguish buying a domain from connecting one. AI prepared a replacement answer and mapped each claim to its source.
The case study contains the actual answer, the inputs and the review. The suggested replacement has not been applied to the pricing FAQ. We have not tested whether customers find it clearer or whether it reduces support questions.
That is the current level of evidence: one draft we can check. I still own the decisions, and we have not proved that AI can run Pineapple independently or grow revenue on its own.
Even the numbers need care. The first draft used a provisional provider snapshot and even labelled its currency incorrectly. We corrected that history and replaced the maintained scorecard with the reconciled Stripe and Paddle reporting source. Presenting that limitation is part of the experiment. An impressive dashboard is no help if its headline means something different from what readers assume.
What a useful result would look like
A customer gets a clearer answer or completes a task more easily. AI produces work we can verify. I spend less time correcting avoidable mistakes. We can explain the cost, the remaining human work, and the failures alongside the successes.
Those are the things we intend to measure. Until we have the evidence, they remain goals.
Try the source-backed FAQ workflow. It includes the Pineapple example, a copyable prompt and checks to use before you consider publishing your own answer.
Updated 1 November 2026.
