Ten founding cohorts
Every tracked product sorted by the year its company was founded, 2016 to 2025 — what each cohort built, how much of it is still here, and why the most recent rows cannot be read at face value.
16 September 2026 · 9 min read · Data frozen 16 September 2026
A catalog of 7,030 products contains something a funding database does not: a population you can follow forward. This report sorts every tracked product by the year its company was founded, then asks what each cohort built, and how much of it is still here.
The short version
- 911 companies in the catalog were founded in 2022, the peak year. 218 of the 2022 and 2023 cohorts are already gone.
- Mortality rises steadily with recency, from 7.1% for 2016 to 12.7% for 2022, then falls sharply for 2025 — which is censoring, not improvement.
- Average opportunity score rises with each recent cohort: 62 for 2020 and 2021, 69 for 2023, 71 for 2024. Newer companies are in better markets.
- The dominant market shifts cleanly by cohort: Design in 2017–18, Finance in 2019, Writing in 2020–21, Image in 2022, Agents in 2023, Coding in 2024.
Coverage, and one distortion you must know about
Founding year is recorded for the large majority of the catalog. Cohorts below cover 2016 to 2025, include both live and shut-down products, and mortality is computed within each cohort against that cohort’s own population.
The 2024 and 2025 rows are shaped by our own collection rather than by the world. In 2026 we added several hundred Singaporean companies in a deliberate sweep, most of them founded in 2024 or 2025. Of 115 live products founded in 2025, 110 are Singaporean. That makes the 2025 row a description of what we went and looked for, not of what was founded globally that year. It is stated here rather than buried because every conclusion about recent cohorts depends on it.
The 2016–2023 cohorts do not carry that distortion and carry the weight of this report.
Finding one: the wave has a shape, and it peaked in 2022
All tracked products, live and shut down, by the founding year of the company behind them. Falcoscan catalog, 16 September 2026.
Founding rises steadily from 2016, accelerates through 2020 and 2021, and peaks in 2022 at 911 companies — the year after the current wave of general-purpose models became publicly usable. 2023 stays close. The decline after that is partly real and partly the lag in any catalog: recently founded companies take time to become visible enough to be tracked.
Finding two: mortality rises with recency, then collapses
Share of each cohort that no longer resolves to a working product. Falcoscan liveness sweep, 16 September 2026.
The 2016 cohort has lost 7.1% of its companies. 2022 has lost 12.7% and 2023 12.1%. The apparent paradox — older companies surviving better than younger ones — is survivorship: a 2016 company that is still listed has already passed through the dangerous years, while a 2022 company is in them now.
The 2025 row reads 0.9%. That is not a cohort that has cracked something; it is a cohort that has not yet had the opportunity to fail. Our mortality report puts the dangerous window at eighteen to twenty-four months after launch, which for the 2025 cohort falls in 2027.
The honest way to read this chart is to look only at cohorts old enough to have been tested, and there the message is consistent: roughly one in eight companies founded at the height of the wave is already gone, against roughly one in fourteen founded five years earlier.
Finding three: newer cohorts are in better markets
Mean opportunity score across each cohort's live products. Falcoscan catalog, 16 September 2026.
Cohorts from 2016 to 2022 sit flat between 62 and 64. Then 2023 jumps to 69 and 2024 to 71. Some of that is the Singapore effect described above, but not all — the 2023 cohort predates that sweep almost entirely, and it is the cohort where the jump begins.
The more likely explanation is that founders got better at market selection as the obvious markets filled. A company started in 2020 could reasonably build a writing tool; a company started in 2023 could see what had happened to writing tools and chose differently.
Finding four: each cohort has a signature market
| Cohort | Founded | Live | Dead | Mortality | Avg opportunity | Largest market |
|---|---|---|---|---|---|---|
| 2016 | 325 | 302 | 23 | 7.1% | 63 | Social |
| 2017 | 348 | 332 | 16 | 4.6% | 63 | Design |
| 2018 | 350 | 328 | 22 | 6.3% | 64 | Design |
| 2019 | 582 | 539 | 43 | 7.4% | 63 | Finance |
| 2020 | 626 | 571 | 55 | 8.8% | 62 | Writing |
| 2021 | 787 | 715 | 72 | 9.1% | 62 | Writing |
| 2022 | 911 | 795 | 116 | 12.7% | 63 | Image |
| 2023 | 845 | 743 | 102 | 12.1% | 69 | Agents |
| 2024 | 330 | 288 | 42 | 12.7% | 71 | Coding |
| 2025 | 116 | 115 | 1 | 0.9% | 70 | Healthcare |
The market with the most live products in each founding cohort, with that cohort's mortality and average opportunity. Falcoscan catalog, 16 September 2026.
Read down the last column and you get a short history of the category: Social in 2016, Design in 2017 and 2018, Finance in 2019, Writing in 2020 and 2021, Image in 2022, Agents in 2023, Coding in 2024. Each is the market that looked most obviously buildable at the moment those companies started.
Two of those are cautionary. The Writing cohorts of 2020–21 and the Image cohort of 2022 entered the two markets that are now the most saturated in the catalog. At the time, both looked like the clearest application of a new capability. That is what the front of a wave looks like from inside it.
The 2024–25 cohort, with its caveat attached
468 live products were founded in 2024 or later. Their market distribution leans toward Coding, Agents and AI Models — infrastructure and autonomy rather than content generation — which is a genuine shift from the 2020–22 pattern.
The ten largest markets among 468 live products founded in 2024 or later. Falcoscan catalog, 16 September 2026.
Heavily shaped by our 2026 Singapore sweep — see the coverage note above. Falcoscan catalog, 16 September 2026.
The country chart is included because leaving it out would be worse than showing it with a warning. It is not a finding about where companies were founded in 2024. It is a picture of where we looked.
Photo: Mark Stebnicki / Pexels. Colour-graded for Falcoscan.
Citing these numbers
Every figure here is from the Falcoscan catalog as it stood on 16 September 2026, and is frozen at that reading. Later changes to the catalog will not alter this page, so a number you quote today will still say the same thing when someone checks it.
Falcoscan, “Ten founding cohorts”, 16 September 2026. https://falcoscan.com/reports/ten-founding-cohorts