№ 00 — MEMBERS ONLY

The Silpo AI Factory playbook: everything obvious is already shipped, and the rubric is not the one on the website

2026-08-13PRIVATE≈ 23 min read

Silpo published two judging rubrics that disagree, and the binding one puts 65% of the score on idea and story against 15% on the MCP engineering everyone will obsess over. But the bigger problem is what a full survey of the company turns up: most of what you would naturally build is already there. Demand forecasting is a signed vendor since 2021. Personalization runs on SAP. An in-app assistant shipped a month before the hackathon. And the idea this playbook recommended first was described in a 2017 interview, which I then reported as if it were true today. This is the teardown with the survey done properly, corrections to my own earlier read included.

WHAT'S INSIDE

01The binding rubric is in the offer, not on the landing page
02The submission bar is far lower than a technical entrant assumes
03Almost everything you would naturally build, Silpo has already built
04Your best idea probably has public prior art, and a juror can find it in one search
05There is no checkout tool, and the industry is retreating from checkout anyway
06Two tools describe a household and are genuinely unadvertised. Receipts are not one of them.
07Silpo already knows more about the household than the tools suggest
08The jury is more findable than I claimed, and one of them pre-empted this article
09Silpo is a profit turnaround with flat stores, and the margin decides which pitches are credible
10The competitor to differentiate from is VARUS, not Instacart
11Base rates say the project dies after the event, and the fix is team composition
12What the prize actually is, and what it explicitly is not
13The rules carry a disqualification chapter nobody will read
14Build the demo against a blackout, because the jury operates in one
15The two weeks, and the seven things a submission must contain
16What a defensible entry looks like once everything taken is subtracted