Supply-chain rollup and the self-improving loop
Reflection on the 2026-05-26 daily thought. Two threads, one shared mechanism between them.
- Thread 1 was the idea of a firm that acquires SMEs captive in a single-customer supply chain (40–60%+ revenue concentration) and replaces the management layer with AI while preserving production. The Italian industrial-district context is the natural home.
- Thread 2 was the YC “self-improving AI-first company” framework — five layers (sensor data / policy / tool / quality gates / learning) — together with the software as temporary artefacts claim that follows from Layer 3.
- The two are not independent: Thread 2 supplies the operational meaning of “managed by AI” that Thread 1 needs to be testable. Without the loop, H6’s prescriptive half is metaphor; with it, the failure modes can be located.
- Landed in the wiki as H6_industrial-ai-rollup-captive-suppliers (new hypothesis, low confidence) plus two concept pages — ai-first-company-loop (the operating model) and software-as-temporary-artefacts (the strategic consequence).
- Author elected borrowed-framework treatment for the YC video (no full source page) — the lighter of the two options the agent proposed.
- Author’s correction to the agent’s first restatement of Layer 5: learning closes the loop into all four upstream layers, including the quality gates themselves — that recursive property is what makes the loop self-improving rather than a static workflow. Captured in ai-first-company-loop inline.
- What stayed as conversation, not wiki: the specific list of plausible Italian dominant-customer anchors (Luxottica, Stellantis, the white-goods industrial-district cluster); family-firm succession dynamics as the supply side of the acquisition pipeline; the bear cases on the artefacts page that were not deeply discussed.