OpenAI opened its Agents API in public beta on September 1016. Coder shipped Claude Code support on its self-hosted Agent Relay14. Microsoft released VS Code 1.140 on September 30, 2026, adding multi-model coordination and multi-folder agent sessions12. Now OutSystems has made its Agent Experience generally available, opening its low-code platform to external AI coding agents including Claude Code, Cursor, Codex, and Kiro while enforcing enterprise governance deterministically7,8. Cognitum One released Ruflo, an open-source multi-agent orchestration platform with 12.5 million npm downloads, adding mission-driven workflows and a single control plane for swarms of more than 100 specialized agents3,4. Google's Android Studio introduced a Bring Your Own Agent feature supporting Claude Code, Codex, and Google's Antigravity13. Independent developers, meanwhile, published orchestration frameworks addressing the same gap: not the agent's capability, but the governance, context, and execution scaffolding around it17,18. The convergence across vendor platforms, IDE releases, open-source harnesses, and practitioner-built tools confirms that "agent control plane" has become a recognized product category spanning every tier of the development stack.
Why it matters
The bottleneck for enterprise adoption of AI coding agents has shifted from model quality to deployment infrastructure. Coder's Agent Relay integration with Claude Code was built because "adoption in regulated industries has been limited by deployment, not demand". OutSystems cited developer survey data showing that 74% of AI-generated code requires significant rework after generation or is never deployed to production at all6. OutSystems says Agent Experience addresses this by letting external AI coding agents shape the architecture while OutSystems generates the code deterministically. Augment Code reported that after AI was already writing nearly 100% of its new code by November 2025, the bottleneck moved to review, verification, planning, incident response, and feedback resolution15. All four cases treat the orchestration-and-governance layer, not the model itself, as the product surface that unlocks new customers and new throughput.
The big picture
The emerging control-plane category now spans at least five tiers of sophistication. At the platform level, Coder's Agent Relay is a self-hosted execution environment for cloud coding agents. It lets Claude Code agents run inside Coder workspaces on the customer's own infrastructure: "network-governed, sandboxed, and fully auditable". Anthropic handles billing and the agent loop; everything the agent touches lives on machines the customer controls.
OpenAI's Agents API takes a different architectural stance. It adds context compaction for long sessions, tool search, parallel programmatic tool calls, and multi-agent support. A Reddit post summarizing the launch frames the key design choice: "The useful split is between orchestration and execution. Teams can reuse the Codex harness without putting every file, secret, or runtime inside OpenAI's sandbox".
OutSystems Agent Experience occupies a distinct position: a governed low-code layer that sits between AI coding agents and production deployment. The system lets agents work at the level of system design while OutSystems handles code generation deterministically. It supports public cloud, private cloud, on-premise, and hybrid deployments. Companies including Yesco, Lowenstein Sandler, and Normal are already building on the platform9. Yesco has compressed its software release cycles to two weeks, and Lowenstein Sandler built a production-ready Litigation Matter Tracker in under three hours.
Cognitum One's Ruflo adds open-source multi-agent orchestration. The platform is "built on a simple premise: an agent is a model plus a harness. The model reasons and writes; the harness supplies the tools, memory, workflows, sandboxes, and controls that let it do real work". Ruflo Mods turn a single mission into an end-to-end agent workflow that researches, builds, tests, validates, secures, benchmarks, and learns.
Google's Android Studio now supports a Bring Your Own Agent feature, letting developers integrate their preferred coding agent and "supercharge it with Android Studio's AI-optimized infrastructure and tool support". Microsoft's VS Code 1.140 introduced HydraFusion, a research preview that tests three workflow paths (Single, Cascade, and Critique) to coordinate multiple models on coding tasks. Experimental remote delegation lets agents discover connected remote hosts and launch sessions based on available resources.
Augment Code documented what happens when the control plane extends across the full software development lifecycle. Between November 2025 and July 2026, size-adjusted output per active developer increased from 12.3 to 55.7, median time to merge fell from 11.2 hours to 3.1 hours, and the company added agents wherever work was piling up rather than automating in lifecycle order.
Below these vendor platforms, individual practitioners are building their own control layers. One developer published AI Workflow Control Plane V2.2, a deterministic control layer that separates task/risk routing, retrieval routing, context assembly, structural analysis, execution, and verification. Another developer built Sanctorum, an Electron desktop app where agents sit at desks on floors with capabilities derived from seat-based permissions; every tool call pauses for user approval by default, and every delegation tree has a spending ceiling.
Between the lines
One practitioner described the core frustration driving this category: "At that point I'm not sure if I'm using an agent or just supervising a very enthusiastic intern"19. A technical article on agent-friendly APIs argued that "the microservices principles everyone stopped talking about five years ago turn out to be exactly what agent traffic needs"1.
ANALYSIS Coder and OpenAI are drawing the orchestration-versus-execution boundary in opposite directions. Coder keeps execution entirely on customer infrastructure while Anthropic retains the agent loop. OpenAI offers a spectrum from its own sandbox to developer-owned environments but hosts the orchestration harness itself. OutSystems adds a third pattern: the platform accepts any agent but generates the production code itself, making deterministic code generation the trust boundary rather than infrastructure location. VS Code 1.140's HydraFusion adds model-selection governance, where the platform decides which model handles which subtask. Cognitum One's Ruflo and the independent projects split differently still, treating the harness itself as the composable, open-source primitive.
What unites them is the premise that the control plane, not the model, is the trust boundary. OutSystems' 74% rework statistic and Coder's regulated-industry framing both point to the same conclusion: without governed orchestration, agent-generated output does not reach production.
What's next
Coder's Agent Relay architecture is designed to onboard additional agent providers beyond Cursor and Claude Code. OpenAI's Agents API remains in public beta with no announced end date. HydraFusion is available to Copilot Pro, Pro+, Business, and Enterprise customers, though Business and Enterprise admins may need to enable the preview. OutSystems Agent Experience is generally available across cloud, hybrid, and on-premise environments, with pre-built claims agents accessible through OutSystems Forge for the insurance sector. Cognitum One's Ruflo is open-source and targets Claude Code, Codex, and other MCP clients5.