Method 1Scanned 2026-10-24
Agent readiness of 100 recent YC startups
We scanned the home pages of 100 companies from Y Combinator’s four most recent batches with the same engine as your report. These are the aggregates; no company is named or scored here.
Scored by method 1. Method 1.1 (28 Oct 2026) fails Named controls on a page whose HTML has no controls, which 8 of these home pages are; we will re-score the sample when the next method ships.
- fail at least one static check; only 3 pass every one
- 97 of 100
- have no machine interface an agent can discover (91 of 97)
- 94%
- have no llms.txt (46 of 97)
- 47%
- is the median static score; half score between 75 and 98
- 88
Method.
- Sample
- 100 active companies from YC’s four most recent batches with more than a handful listed (Fall, Summer, Spring and Winter 2026), from the public yc-oss directory dataset as of 2026-10-24.
- Sectors
- 25 each from YC’s own classification: B2B software, developer tools, fintech and consumer. Industrials, healthcare, real estate, government and education are not in the sample.
- Draw
- Within each sector, a round-robin over the four batches, ordered by a SHA-256 of a fixed seed and the company’s slug: the same data and seed give the same sample.
- Engine
- Themelior’s static checks (Find, Act, Safe), method 1, one scan of each company’s listed home page as ThemeliorBot, one site at a time, two seconds apart. No real agent runs.
- Correction
- The first pass flagged a CAPTCHA on 3 home pages; all 3 were false positives of a rule that matched widget names in the page text, since replaced (see the method changes). Every home page was fetched once more and that sub-check recomputed with the markup rule.
Most common findings.
All 100 sites answered; 3 served a bot challenge instead of the home page, so shares are of the 97 readable home pages.
| Finding | Sites of 97 |
|---|---|
| No machine interface (no OpenAPI or MCP description) | 91 |
| No llms.txt | 46 |
| No JSON-LD structured data | 34 |
| Broken heading outline | 34 |
| No main landmark | 21 |
| No sitemap | 17 |
| Controls an agent cannot name | 17 |
| An empty HTML shell until JavaScript runs | 8 |
By sector.
Groups of 24 or 25 readable sites: read the differences as direction, not proof.
| Sector | Readable | Median | llms.txt | JSON-LD | Machine interface |
|---|---|---|---|---|---|
| B2B software | 24 | 90 | 54% | 71% | 8% |
| Developer tools | 25 | 93 | 68% | 80% | 0% |
| Fintech | 24 | 88 | 46% | 63% | 4% |
| Consumer | 24 | 79.5 | 42% | 46% | 13% |
Limits.
- Static checks only: what a home page offers, not whether agents finish tasks on it.
- One home page, one scan, one moment (2026-10-24); deeper pages and later deploys can differ.
- No JavaScript and one user agent: sites may treat other agents or a browser differently.
- Not a random draw from the web: recent YC companies are early-stage and AI-heavy, so this is probably an optimistic picture of the web at large.
- With 97 sites, a share near 50% carries about ±10 points of sampling error (95%); a sector’s share, about ±20.
Where does your site stand?
Scan your own site