
Which AI tools ship an API
An API is the least glamorous thing an AI product can ship and one of the most revealing. Across 6,225 live products where we know the answer, 60.1% have one — and they fail noticeably less often.
16 September 2026 · 8 min read · Data frozen 16 September 2026
An API is the least glamorous thing an AI product can ship and one of the most revealing. It says the company expects to be part of someone else’s system rather than the whole of it. Across 6,225 live products where we know the answer, 60.1% ship one — and the ones that do are noticeably less likely to disappear.
The short version
- 3,742 of 6,225 live products with a known answer ship an API — 60.1%.
- Products with an API fail at 7.4%. Products without fail at 10.5%. That is a 42% higher failure rate for the ones with no programmatic access.
- The market range is extreme: 95.6% in Security, 14.5% in Writing.
- Bootstrapped companies ship APIs least (26.9%). Venture-funded companies ship them at roughly 70% regardless of stage.
Coverage and definition
“Ships an API” means a documented programmatic interface intended for third parties — not an internal endpoint, and not a webhook-only integration. 6,225 of 6,441 live products have a recorded answer; 216 do not and are excluded from every rate below rather than assumed to be false.
Finding one: an API goes with surviving
Share of tracked products in each group that no longer resolve to a working product. Falcoscan liveness sweep, 16 September 2026.
7.4% against 10.5%. This is one of the larger survival differences we have found between two groups defined by a single product property, and it is worth being careful about what it means.
We do not think shipping an API keeps a company alive. The likelier reading is that an API is a marker for two things that do: a technical buyer, and integration into a customer’s workflow. A product reached through code is a product someone has spent engineering time wiring in, and that time is a switching cost. A product reached only through its own interface can be abandoned in an afternoon.
There is also a survivorship caution attached. Building an API takes resources, so better-resourced companies are more likely to have one, and better-resourced companies survive more often for reasons that have nothing to do with the API. The honest claim is correlation with a plausible mechanism, not proof of cause.
Finding two: a six-fold spread by market
Percentage of live products with a known answer that ship an API. Falcoscan catalog, 16 September 2026.
Security leads at 95.6%, because a security product that cannot be automated is not a security product — it has to run continuously against systems nobody is watching. AI Models (91.9%) is definitional: the API is the product.
Writing sits last at 14.5%, with Learning (17.6%) and Design (26.4%) just above. Those are markets where the output is consumed by a person, one piece at a time, and where there is no obvious system to plug into. It is also, precisely, the set of markets our wrapper-risk report identifies as thinnest — a product that is a prompt with an interface has little to expose programmatically that the underlying model does not already expose better.
Finding three: it is a funding question too
Live products with a known answer at each stage. Pre-seed has a small known population and is shown for completeness. Falcoscan catalog, 16 September 2026.
Series A (71.3%) and growth-stage (71.1%) companies ship APIs at essentially identical rates. Bootstrapped companies ship them at 26.9% — less than half. An API is real engineering that produces no visible product, which is exactly the kind of work an unfunded team defers.
Read alongside the survival data, that compounds a disadvantage the bootstrapping picture already describes: unfunded companies cluster in the most contested markets and are also least likely to build the thing most associated with staying embedded.
One caveat on that chart. Our pre-seed group has a known-answer population of just 92 products and shows 100%, which is a coverage artefact rather than a finding — for recently added companies we tend to record the API field only when we have positively confirmed it. Read the four large stages.
What shipping one looks like
The highest-opportunity products in the catalog that expose a public API.
Every market
| Market | With API | Known | Rate |
|---|---|---|---|
| Security | 237 | 248 | 95.6% |
| AI Models | 284 | 309 | 91.9% |
| Automation | 90 | 100 | 90% |
| Support | 202 | 225 | 89.8% |
| Finance | 234 | 262 | 89.3% |
| Data | 255 | 287 | 88.9% |
| Human Resources | 75 | 87 | 86.2% |
| Healthcare | 98 | 114 | 86% |
| Marketing | 229 | 274 | 83.6% |
| Sales | 225 | 270 | 83.3% |
| Agents | 269 | 329 | 81.8% |
| Ecommerce | 227 | 292 | 77.7% |
| 3D & AR/VR | 45 | 65 | 69.2% |
| Gaming | 38 | 56 | 67.9% |
| Voice | 160 | 268 | 59.7% |
| Legal | 44 | 77 | 57.1% |
| Research | 45 | 79 | 57% |
| Coding | 225 | 419 | 53.7% |
| Real Estate | 39 | 77 | 50.6% |
| Search & SEO | 32 | 67 | 47.8% |
| Avatars | 30 | 66 | 45.5% |
| Image | 157 | 391 | 40.2% |
| Social | 100 | 264 | 37.9% |
| Video | 102 | 292 | 34.9% |
| Music | 20 | 60 | 33.3% |
| Productivity | 92 | 295 | 31.2% |
| Design | 94 | 356 | 26.4% |
| Learning | 43 | 245 | 17.6% |
| Writing | 51 | 351 | 14.5% |
Photo: Brett Sayles / 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, “Which AI tools ship an API”, 16 September 2026. https://falcoscan.com/reports/which-ai-tools-ship-an-api