Reynar: three 2001–2005 machine-readable-knowledge projects died on schema design, population, and maintenance — and only maintenance has a credible 2026 fix

semantic-webontologyknowledge-graphcodificationai-strategyagentic-aiconflict-of-interest

Jeff Reynar — who built Microsoft's Smart Tags feature (2001), watched SharePoint's two 2001 launches diverge on the same axis, and later ran Google's DAFFIE question-answering knowledge base (2005) — argues in a 2026 retrospective that all three projects failed on the same three sub-problems: schema design, population, and maintenance. He argues the common explanation ("too hard, too academic") doesn't hold, since SQL was by his account harder and won completely; what the three attempts actually lacked was an answer to the ongoing cost of upkeep. His claim about 2026's difference: an AI agent can now do that maintenance "as a byproduct of the work," rather than as "a job somebody has to remember to do" — closing the one sub-problem of the three that, in his account, never had a working answer before.

Caveat, load-bearing to how much this claim can carry: Reynar frames the three projects as independent attempts, but two of them (Smart Tags, SharePoint) are the same company (Microsoft), leaving effectively two organizational contexts rather than three. The whole account rests on one person's memory, written twenty-plus years after the earliest of the events, with no contemporaneous record corroborating it found this session. Reynar also discloses he is building a company premised on exactly the 2026 fix he's describing — his diagnosis of why the earlier attempts failed is the more falsifiable, less self-interested part of the piece than his optimism about agents-as-maintainers.

This is the historical throughline behind claim-shadbolt-hall-berners-lee-2006-name-ontology-maintenance-cost-as-often-quoted-concern and the entity entity-jeff-reynar; see claim-tung-roussiere-2026-codify-tacit-knowledge-into-private-knowledge-graphs-as-moat for the 2026 pitch this retrospective is implicitly addressing.

Correction history.

• 2026-10-15 — this note's caveat says "Reynar frames the three projects as independent attempts." A closer read of the same primary (sha 3df4edd9…) found at promotion of 10-inbox/raw/2026-10-14-jeff-reynar-built-smart-tags-at-microsoft-ran.md shows Reynar's own language never claims independence: he introduces them as "three large efforts" and explicitly foregrounds, rather than conceals, that two are "two bets, same company, same year" — he builds his central argument on that intra-company comparison (Portal Server vs. Team Services), not on treating the three as statistically independent data points. The word "independent" here is the vault's own interpretive gloss, not a paraphrase of Reynar's claim. This does not resolve question-corroborate-reynar-three-knowledge-projects-schema-population-maintenance — the account is still effectively two organizational contexts and one person's memory — but the caveat above should be read as flagging a vault-introduced overstatement, not an overstatement in the source itself.