Build vs Buy
Reviewed 6 Oct 2026. Verdicts and tallies recompute from the tiles on every build, so this tab carries no date of its own.
Build, buy or wait, at each level, read from the firm side: pick the level a firm is at (or aiming at) and see the verdict, the lab capabilities pressing on it, and the Landscape tools at that level.
The verdict, in one line. Buy the commodity layer (search, retrieval, productivity - a bought tool does most of the work up to L3-L4); build the thing that is your edge (the signal architecture, the eval infrastructure and the workflow that wraps the model). "The model can be bought, the workflow has to be built." The same line runs through a venture house's 2024-2026 thesis in other words - the durable product is the cognitive architecture around the model, and a copilot that sells the tool is a different business from an autopilot that sells the work; on this ladder that is the L4-to-L6 step. Below, pick a level to see the specific verdict, the lab capabilities pressing on it, and the tools that fit.
As of v90, 21 Oct 2026. Tallies computing from the tiles... What changed this build: no build↔buy migration and no verdict changed - no firm's level moved. The Maturity Matrix now places every firm from its tile, quant and HFT read as separate archetypes, and these tallies are computed from the tiles at load rather than typed.
Start here: where is the firm? Pick a level
The four-question test
The general rule, across levels
Buy at L1-L3 (the tools are commodities and the vendor market is deep). Build selectively at L4, where vendor tools hit their ceiling. Build at L5+, where the reorganised process itself is the edge and there is almost nothing to buy. Wait only where the lab roadmap is about to commoditise what you would build.
Why: below L4 you are renting general-purpose capability every competitor can rent; at L5+ the moat is domain-trained models and redesigned workflow, which cannot be licensed. Read the per-level verdict above for the specifics.