Urban Pousses — Competitive intelligence and niche arbitration by AI agents

Urban Pousses grows microgreens indoors in the Oise and sells them to chefs, local authorities and regional distributors. It is a niche business, and that niche has an awkward feature: it exists in no public statistics. Neither Agreste, nor FranceAgriMer, nor the market bulletins track microgreens. The only market figures available come from private research firms that copy one another and publish the same growth rate from one edition to the next on different baselines.

The result is a grower operating blind — with no view of who produces what, how far away, at what price, and above all no way to tell which of their growth ideas are realistic. Should they sell seeds? Train other growers? Sell turnkey containers? Go after school canteens? Each of those questions represents an investment, and none of them had a documented answer.

I built a system of five research agents launched in parallel, each on one market segment: regional microgreens growers, industrial-scale indoor farming players, equipment and know-how suppliers, consumer market and distribution, and finally market context and trends. Every company cited is verified against the public company register — legal form, incorporation date, headcount, trading status — and every statement is flagged as either read at source or inferred.

That protocol produced two findings no keyword monitor would ever have surfaced. First, the market is not competitive but made up of regional near-monopolies, because the product does not travel: almost every verified grower delivers to its own catchment area only. Second, the Oise, the Somme and the Aisne are a blank spot — no active microgreens grower was identified there, cross-checked against the county's own directory of market gardeners. The strategic question was therefore not "how do we differentiate" but "how do we occupy an already empty territory before it closes".