You are putting together a shortlist of companies whose APIs an agent could call, and someone hands you Greylock Partners' portfolio as a place to start. It looks promising. Greylock has 157 portfolio companies on the apis.io network, and you want to know quickly which ones are worth your time.
The first number you meet is a summary. Of those 157, 91 (58%) publish APIs that apis.io indexes. Those 91 average 37.8 on the Kin Score and 24.2 on Agent Readiness, which puts the portfolio in the thin band. If you stopped there, you might decide the whole book is a weak place to look.
The post warns you not to stop there, because that average hides a split. Of the 91, 57 have an API count on a published basis, meaning the count rests on what the company itself published. Those 57 average 47.3 on the Kin Score and 33.8 on agents. The other 34 have a count that apis.io arrived at by splitting what the company publishes, and they average 24.4 and 10.3. Every one of the portfolio's 21 human-only providers sits in that second group.
Now you look at the top of the book. Anthropic leads at 88.8, with an agent score of 46.7 and six APIs. OpenAI is at 84.4 with 57.1 on agents and 19 APIs. LinkedIn scores 74.8 and 41.9 across 17 APIs, Okta 74.3 and 53.0 across three, and Revolut 70.0 and 53.9 across eight. In all there are eight exemplars and eight strong providers, followed by 26 developing, 28 thin, 16 emerging and 5 minimal.
The names you might have expected to rank highest are not the ones to watch. Lightfield has a single API and the best agent score in the portfolio, 65.6. Common Room scores 63.8 on agents from three APIs. Cloudera has the most APIs in the book, 38, and is only agent-aware at 27.4. The post's reading is plain: in this portfolio, breadth does not predict agent readiness, so a long API list should not make you assume a company is ready for agents.
Then you reach the tail, and it is where an agent builder might feel the gap. The 21 human-only providers include companies you could plausibly want to call. Wiz scores 14.6 with an agent score of 0.0. Roblox scores 28.2 across eight APIs, with 3.1 on agents. StackBlitz, Domo, Abnormal and Clockwise sit at 0.0 on both scores. Modular and Reduct score in the forties on the Kin Score but only 2.5 on agents. The post reads that pattern as documentation that exists without a contract behind it.
This is where the post turns on its own record. A split API count, it says, is not a penalty and says nothing about product quality. It means the portfolio company did not hand apis.io one contract per API, so apis.io had to cut the surface up itself. Those split counts are apis.io's work. Greylock's companies should not be credited for them, and they are not blamed for them either. They score low because there is less first-party material for the rubric to read. So when you see 24.4 against 47.3, you are partly looking at how the catalog had to build its own numbers.
There is also context the post offers. Greylock invests at pre-seed, seed and Series A, so the thin average partly reflects how young the book is. The AI companies at the top show that a young company can score well when it ships a contract early.
What you take away is a better way to read the list. Do not rely on the portfolio average of 37.8. Check whether a company's count is published or split before you trust its score. For the 34 split-basis companies, the cheapest fix the post names is one published OpenAPI per product. And for the other 66 companies with no published API on the network, the post says no number would be honest until they publish one, so you should not treat their absence as a score at all.
Read the original: https://apis.io/2026/10/27/greylock-portfolio-splits-in-two-at-the-published-contract/