Y Combinator Companies Dataset

What’s in the file

37 columns over 6,194 rows. Distinct counts and null rates are measured from the file itself, not estimated.

Things worth noticing

• 214 companies are still listed Active and their website did not answer when we checked it. The directory records what a company last told YC; website_live records what was true on the day of the check, and the gap between the two is the most useful thing in this file.
• A domain that now lands somewhere else is the clearest public tell for an acquisition. 22.5% of Acquired companies redirect off their own domain against 4.6% of Active ones — Pincites now resolves to filevine.com, Dashdive to rhombus.com.
• Survival decays with age about as steeply as you would guess and further than most people expect: 21.2% of Winter 2012 is still operating, against 79.1% of the 2022 batches. Some of that gap is simply time — the recent cohorts have not had the years in which to fail — which is why batch_active_rate ships as a column you can interrogate rather than a claim.
• AI went from a niche tag to nearly half the intake in nine years: 7.9% of the 2015 batches carry one, against 45.4% of 2024. is_ai is computed from an explicit list of eight tags that actually occur in the source, so the definition is auditable instead of a keyword guess.

Put an agent on it

Anything you were about to do with this file by hand is a job that runs every week forever. Each of these is a working prompt — open it and it is already loaded.

Every month, compare the current Y Combinator company directory against the snapshot from last month.

Report only what changed:
- Companies whose status moved (Active to Acquired, Active to Inactive, or into Public)
- Companies whose team_size grew or shrank by more than 30%
- New companies added, grouped by batch and industry
- Any company in the batches I follow that started or stopped hiring

For each change, give the company name, its batch, the old and new values, and its YC profile link. Sort by how surprising the change is, not alphabetically. If a month is quiet, say so in one line rather than padding the report.

Every week, request each company website in this dataset and record the HTTP status and the domain the request finally lands on.

Flag a company when:
- Its site stopped resolving entirely, having resolved on the previous run
- Its site now returns 4xx or 5xx for two consecutive runs
- Its domain now redirects to a different registrable domain than before

For each flag, report the company, batch, its status as the directory still records it, the previous and current HTTP status, and the redirect target where there is one. Call out separately any company still listed Active whose site has been unreachable for three runs — that gap between recorded status and observed reality is the whole point of the check.

I will give you a thesis in one sentence — for example "seed-stage B2B infrastructure companies outside the US that are still small enough to partner with".

Filter this dataset to the companies that match it. Use the structured columns to do the work: batch_year and years_since_batch for stage, team_size and team_size_bucket for size, industry, subindustry and tags for the space, hq_country and remote_policy for geography, and is_operating plus website_live to drop the ones that are gone.

Return a ranked table of at most 25 companies with name, batch, team size, location, tags and YC profile link. After the table, write three sentences on which filters did the most work and which companies are borderline and why. Do not pad the list to 25 if fewer genuinely match.

Load it yourself

The file is a plain parquet. Nothing here is specific to us.

-- No download step: DuckDB reads the parquet over HTTP range requests.
INSTALL httpfs;
LOAD httpfs;

CREATE VIEW y_combinator_companies AS
SELECT * FROM read_parquet('https://cotera.co/datasets/y-combinator-companies.parquet');

SELECT * FROM y_combinator_companies ORDER BY company_id LIMIT 20;

More datasets