Welcome to Harvest
Why product intelligence is fragmented, why AI changes that, and what we're building to make the new pattern stick.
Paul Pounder
2 May 2026 5 min read
Product teams already write everything down. Sales notes, support tickets, customer-research transcripts, competitor scans, decision memos, retro outcomes — every team in every product org generates more written evidence in a quarter than any human can read in a year.
The problem is that none of it is connected. The strategist updates a deck, the support lead writes a Linear ticket, the researcher publishes a Notion page, the salesperson drops notes in HubSpot. Each artefact is an island. Nobody holds the whole picture.
Why this used to be unsolvable
Synthesising fragmented evidence into a living model of "what we know about our product right now" has always been theoretically possible and practically impossible. The cost of reading everything, summarising everything, cross-referencing everything against last quarter's beliefs, and writing down what changed was just too high.
So teams kept the cost down by never doing it. We have product roadmaps based on the loudest voices. We have personas written once and never updated. We have strategy decks that age out the moment the next research insight lands.
Why AI changes the economics
A capable language model can ingest a thousand pieces of evidence in an afternoon, identify which insights reinforce or contradict prior beliefs, and propose updates to a structured knowledge base — without forgetting what was true last week. The cost curve flips.
That doesn't mean LLMs can replace product judgment. They can compile, structure, and cross-reference; they can't decide what to ship. The pattern that works splits the labour explicitly: humans set foundations, the LLM maintains the wiki, humans make decisions, the LLM reconciles outcomes back into the baseline.
What we're building
Winnow is the open-source reference implementation of Harvest, and productharvest.org is the home for the methodology behind it.
Self-host today; hosted version on the way. Either way, the wiki is yours, the evidence is yours, and every recommendation traces back to the source it came from.
More to follow as the pattern hits real product orgs and we learn what bends.