LaunchDarkly has launched AgentControl for software program groups managing AI brokers in manufacturing, extending its platform into AI operations.
AgentControl is designed to let groups change how an AI agent behaves with out redeploying the underlying utility. The system can alter an agent’s runtime behaviour in lower than 200 milliseconds, together with switching to a different mannequin or triggering a fallback response earlier than a consumer sees an unsatisfactory reply.
The launch displays a broader shift in software program growth as companies transfer AI brokers from experimentation into dwell, customer-facing providers. That transition has created operational points that differ from these in typical software program releases, as a result of agent behaviour can change in keeping with prompts, fashions and manufacturing context even when the core utility code stays the identical.
Governance and operational oversight
As soon as deployed, brokers’ settings and administration processes usually sit throughout a number of groups and frameworks. That fragmentation has elevated demand for shared guidelines on governance, model management and launch administration, significantly the place engineering groups have to intervene rapidly as dwell circumstances change.
AgentControl combines runtime intervention with a broader operational layer for groups operating AI brokers. It presents central oversight throughout groups and techniques, benchmarking earlier than adjustments attain prospects, progressive roll-outs, monitoring with trace-level visibility, and the flexibility to regulate brokers primarily based on manufacturing knowledge.
LaunchDarkly constructed its identify in function administration and software program launch controls, giving engineering groups a solution to alter software program behaviour in manufacturing with no full redeployment. With AgentControl, it’s making use of that method to AI techniques, the place intervention could also be wanted throughout a dialog slightly than over a traditional deployment cycle.
Addressing AI reliability considerations
That issues as a result of AI brokers can produce inconsistent outputs, whereas mannequin efficiency might shift over time. Companies utilizing these techniques have been on the lookout for methods to observe such adjustments, check updates extra fastidiously and scale back the chance of errors reaching finish customers.
In apply, the product is meant to cowl each direct management and broader oversight. Groups can change configuration at runtime, monitor agent efficiency dwell and set requirements for the way agentic workflows are ruled throughout completely different components of an organisation.
Cameron Etezadi, Chief Expertise Officer at LaunchDarkly, mentioned the corporate noticed clear overlap between established software program management issues and the operational challenges now rising round AI brokers.
“LaunchDarkly has at all times been about giving software program groups management at runtime over what their software program does in manufacturing,” mentioned Etezadi.
“The toughest issues in AI, like mannequin drift, unpredictable outputs, and the lack to intervene quick sufficient, develop into precisely the issues our platform was constructed to resolve. We did not should reinvent the platform; we simply needed to lengthen it to satisfy the calls for of an AI SDLC and agentic-driven workflows,” added Etezadi.
Business help for runtime controls
The corporate additionally pointed to help from Cursor, which develops AI coding instruments utilized by software program groups. Each companies are positioning runtime controls as an more and more necessary a part of AI software program supply as extra organisations transfer from pilot tasks to manufacturing deployments.
“Cursor is how the world’s main enterprises are constructing with AI. As extra AI-powered merchandise and agentic capabilities attain manufacturing, runtime management turns into important infrastructure alongside the event workflows and controls groups already belief,” mentioned Brian McCarthy, President of International Income and Subject Operations at Cursor.
“LaunchDarkly constructed an extra layer for that surroundings, and AgentControl extends it to the agent lifecycle in a means that enhances how Cursor’s prospects already construct,” added McCarthy.
LaunchDarkly says it serves 1000’s of shoppers worldwide, together with 1 / 4 of the Fortune 500, from its base in Oakland.