04 — Product · PR-940

LaunchDarkly

A hosted service for changing what software does in production without a redeploy: features and AI agent settings are switched per audience at runtime, released in monitored stages that halt or reverse when metrics worsen, and tested as experiments, with errors and user sessions tied back to the change that caused them. Free entry plan; paid plans priced by usage.

// FIG. 04 · Capability coverage

Each tile is an L1; the bar is how many of this product's capabilities realize something beneath it. Our judgment, not a vendor claim.

Direct & Govern

Finance

People & Workplace

Customer & Market

Products & Services

Operations & Supply

Technology & Data

Resilience & Responsibility

Industry L1s

Product capabilities · 6

How to read maturity and confidence

Maturity · how much it covers

core: What the product is built around; the capability is covered end to end.
strong: Covered well, with a few gaps a buyer would notice.
partial: Covers part of the capability; expect to pair it with something else.
emerging: New or thin; on the roadmap more than in the product.

Confidence · how well it is evidenced

Evidence confidence: high: The vendor's own documentation describes this module doing it.
Evidence confidence: medium: Documented, but how far it reaches into the capability is our reading.
Evidence confidence: low: Our judgment; we found no public page that confirms it.

PC-940.10
Feature Flags
Evidence confidence: highMaturity: core
Separates deploying code from releasing it, so a team can ship a feature dark, show it to chosen users or segments at runtime and switch it off at once if it misbehaves. Sign-offs and an audit trail govern each change, and flags nobody uses are found and retired.
Vendor page: launchdarkly.com/how-it-works/feature-flags/ ↗
PC-940.20
Release Autopilot
Evidence confidence: highMaturity: emerging
Automates the path from merged code to full release. Each pull request is rated for risk and comes with a suggested flag and release plan, and the agreed run of environments, sign-offs and checks starts by itself once the build is deployed.
Vendor page: launchdarkly.com/how-it-works/release-autopilot/ ↗
PC-940.30
Guarded Rollouts
Evidence confidence: highMaturity: strong
Lets a release protect itself, since exposure grows in steps only while the new version holds up against the old on the chosen measures, and a clear regression alerts the owner or reverses the change.
Vendor page: launchdarkly.com/docs/home/releases/guarded-rollouts ↗
PC-940.40
Experimentation
Evidence confidence: highMaturity: strong
Experiments attached to flags, from A/A checks and A/B/n tests to multi-armed bandits, read out with Bayesian or frequentist statistics and able to use metrics computed in the customer's own data warehouse.
Vendor page: launchdarkly.com/docs/home/experimentation ↗
PC-940.50
Observability
Evidence confidence: highMaturity: partial
Shows what a release did to real users by tying errors, logs, traces, session replays and heatmaps to the version that was live, and notifies the team in chat or through its paging tool when a release crosses a set threshold.
Vendor page: launchdarkly.com/how-it-works/observability/ ↗
PC-940.60
AgentControl
Evidence confidence: highMaturity: emerging
Lets teams change how an AI agent behaves in production without a code release. Prompts, model and tool settings are runtime configuration, and each change is graded by evaluator models, released to a growing share of users and watched for cost and response time.
Vendor page: launchdarkly.com/platform/agent-control/ ↗

Sources · checked 2026-11-06

The public vendor documentation this mapping is based on. Maturity and confidence are our judgment, not the vendor's claim.

How it works - Feature flags | LaunchDarkly launchdarkly.com/how-it-works/feature-flags/
Release Autopilot | LaunchDarkly launchdarkly.com/how-it-works/release-autopilot/
How it works - Experimentation | LaunchDarkly launchdarkly.com/how-it-works/experimentation/