The Feature Flags tag is small — 28 providers tagged, 54 APIs — and it is one of the strongest corners of the catalog. Its members average a Kin Score of 50.2 against 20.2 for the rest of the catalog, and an Agent Readiness of 38.4 against 11.9. Not one of the 28 sits in the bottom two bands. And yet the thing this category sells, the ability to turn a change off again, is the thing its APIs are least likely to tell an agent how to do: only 39.3% of feature flag providers document reversibility.
The five the tag page ranks first are all exemplars. Kinde leads at 88.8, an identity company that ships flags alongside authentication, which is why it repeats here after leading our authentication showcase; its score is carried by access clarity at 93.4 and developer ergonomics at 91.1, and it is the only one of the five with idempotency lit. LaunchDarkly is second at 85.5 with a perfect 100.0 on access clarity, 68 API pages and the widest agent surface in the group: dry-run mode, delegated identity, protected resource metadata and documented reversibility all lit. Flagsmith, the open source option, scores 79.4 on 24 API pages, strong on access clarity at 93.4 and softer on contract quality at 59.9. Optimizely scores 77.7 across 414 API pages, the largest surface here, leaning on discoverability at 90.8 while access clarity trails at 60.5. Statsig rounds out the five at 76.8, and with LaunchDarkly it is one of only two in the group that lights dry-run mode.
Against the rest of the market, this tag is a different population. The cohort median is 46.3 against 13.4 for the rest of the catalog. A quarter of the members, 25.0%, are Kin Score exemplars, against 1.6% of everyone else; 0.0% are emerging or minimal, against 22.7% and 41.0% of the rest. Contracts are nearly universal — 92.9% have a published spec against 32.6% — and 57.1% ship OpenAPI examples against 15.8%. The weakest point in the band ladder is the long tail: 42.9% of the tag is still thin, so the gap between the five leaders and the bottom twelve is the real story inside the category. The soft spot even among the leaders is contract governance, which runs from 31.8 to 45.5 across all five.
On agent readiness, feature flags is built for agents more than almost any category we have measured. 64.3% of members are agent-ready against 14.2% of the rest of the catalog, and 3.6% are human-only against 53.9%. The lit-rate gaps are wide: agentic access at 82.1% against 22.3%, an MCP server for 60.7% of members against 14.1%, auth clarity at 92.9% against 38.2%, rate limit signals at 75.0% against 30.4%. Delegated identity and protected resource metadata are each lit for 46.4% of members, more than three times the market. Where it goes dark is where the rest of the market is also dark — agent cards at 0.0%, agentic commerce at 0.0%, well-known catalogs at 7.1% against 8.9% elsewhere, one of the few dimensions where this tag trails.
The finding is about the kill switch. A feature flag is reversibility as a product: flip it on, watch, flip it off. That should make this the category where every write is documented as undoable, safely retryable and testable before it runs. Instead reversibility is documented by 39.3% of members, idempotency by 21.4%, and dry-run mode by 14.3%. Those are all several times the market rate, and they are still a minority. Four of the five leaders leave idempotency unlit. An agent toggling a production flag through one of these APIs will find the auth well explained and the rate limits well signalled, and will then have to guess whether a retried toggle double-applies and how to put the old state back. The category that taught the industry to ship safely has not yet written that safety into its own contracts.