Not every corner of the AI market dies at the same rate. We rank every category in our catalogue by the share of its tools that have since shut down or been acquired and sunset — measured against the tools we actually tracked, not estimated from the outside.
Where a tool sits matters more than the market-wide average. The safest category we track is Engineering & Simulation at 2.1%; the riskiest is Coding Assistants at 16.7% — 8 times likelier to be dead. Across the whole catalogue the rate is 13.9%.
Among major categories (50+ tracked tools), Coding Assistants leads at 16.7% — 9 of 54 gone. Smaller categories can post higher rates off a handful of deaths, so read the sample column alongside the rate.
Categories with fewer than 12 tracked tools are excluded: one death in a category of three is a 33% rate that means nothing. Rates are calculated against every tool we have ever catalogued in that category, including the ones still running.
ToolDirectory.AI, "Which AI Tool Categories Fail Most" (as of November 13, 2026): AI tool failure rates range from 2.1% (Engineering & Simulation) to 16.7% (Coding Assistants) across 40 categories — a 8× spread. Among categories of 50+ tracked tools, Coding Assistants is highest at 16.7%. https://tooldirectory.ai/research/ai-category-failure-rates
Free to quote with attribution under CC BY 4.0. Figures move as the dataset grows — please cite the date alongside the number.
Coding Assistants, at a measured 16.7% failure rate — 9 of the 54 Coding Assistants tools ToolDirectory.AI has catalogued are gone, as of November 13, 2026. Among larger categories (50+ tracked tools), Coding Assistants is highest at 16.7%. Every ranked category clears a floor of 12 tracked tools, but read the sample size alongside the rate: smaller categories move further on a single shutdown.
Substantially — from 2.1% (Engineering & Simulation) to 16.7% (Coding Assistants) across the 40 categories we rank, a 8× spread. The whole-catalogue rate is 13.9%, so a market-wide average understates the risk in the worst categories and overstates it in the safest.
Coding Assistants (16.7%), AI Content Writing (12.4%), BI & Analytics (11.7%), Developer Tools (11.3%), AI Art & Image Creation (10.3%) — ranked by the share of catalogued tools that are now gone, across categories with at least 50 tracked tools. The full table on this page ranks every category down to a floor of 12.
Dead tools in a category divided by every tool we have ever catalogued in that category, at any lifecycle. A tool counts as dead if it shut down, or was acquired and had its brand retired; acquired tools still shipping under their own name are excluded from the numerator. Categories with fewer than 12 tracked tools are excluded entirely, because rates on tiny samples are noise.
No. Commonly cited AI startup failure figures (80–90%) are company-level estimates. This is measured at the product level inside a fixed catalogue: of 2,887 AI tools tracked, 401 (13.9%) are gone. The per-category rates on this page break that number down by what the tool actually does.
The pattern in our data is concentration risk: categories sitting closest to a foundation-model vendor's own roadmap lose tools fastest, either absorbed by acquisition or undercut by a native feature. Categories built on workflow and integration depth retain tools longer. We publish the rates rather than the theory — the underlying dataset is public at /feed/graveyard.json.
It is computed live from the Graveyard on every rebuild, so it is current as of November 13, 2026. Every entry is hand-reviewed, sourced and dated by an editor rather than auto-scraped.
One email each Friday. The week's additions, the week's deaths, and one thing we changed our mind about. No drip sequences, no AI-generated filler.
Receive weekly updates so you can stay up-to-date with the world of AI
Receive weekly updates so you can stay up-to-date with the world of AI
The AI tools directory for discovering, exploring, and comparing the most innovative AI tools in the industry