Everyone Is Now Selling You The Only Control Plane For AI Agents
Updated: 5 days ago
Pattern — Workato's three July moves and the land grab underneath them, 23 July 2026
Workato shipped three things in fifteen days that look like product news and are actually one repositioning. It is trying to stop being a vendor of agents and start being the layer that governs everybody's agents — including its competitors'. That is a rational read of where durable value sits once agents commoditise.

It is also a race Workato entered after Salesforce, against a field that includes the API-gateway incumbents and the hyperscalers, and buyers are about to be pitched four mutually exclusive "single control planes" in the same quarter.
The Three Signals
9 July — Headless API and Agent Guardrails. Workato's Genies can now be embedded into any application surface — customer-facing products, internal tools, or another agent's environment — inheriting governance from Workato Enterprise MCP. Every call carries the calling identity, access is scoped per agent, revocation is instant via key rotation, and PII is blocked, redacted or tokenised before it reaches the model. The agent stops being a destination you visit and becomes something that travels, with the policy attached.
16 July — Enterprise MCP Registry. A governed system of record for MCP servers, alongside a gateway, a proxy and a catalogue of more than sixty production-ready servers. The load-bearing sentence is the scope: the registry covers every MCP server across the enterprise, whether built by Workato, AI vendors, software providers, partners or internal development teams. Workato is volunteering to govern its rivals' agents. Verified User Access ties every agent action to the requesting user's identity rather than a shared service credential, with audit trails recording who initiated it, which agent ran it and which systems it touched.
23 July — FinShield AI with L&T Technology Services. A fraud-investigation agent for banks, built by a systems integrator and published on Workato's marketplace. Workato wrote none of the domain logic. It supplied the governed execution layer and the shelf.
What They Add Up To
Read individually, three ordinary launches. Read together, a deliberate withdrawal from the part of the stack that is about to get cheap.
The agent itself is becoming a commodity. Any model vendor can build one, and increasingly does. What does not commoditise is the boring, expensive, politically fraught work underneath: which identity an action runs as, which systems it may touch, what happened and in what order, and who signs off. That is the position Workato is claiming — and the FinShield launch shows the operating model, with the vertical expertise pushed out to a partner while the platform keeps identity, policy and audit.
The timing is not an accident. In the same fortnight, OpenAI launched a deployed enterprise agent product with its own forward-deployed engineers and systems integrators. Model vendors are moving up into applications. Orchestration vendors are moving down into infrastructure. Both are running from the same middle.
Why This Is Harder Than It Looks
Workato's problem is that the position it is claiming is already claimed, louder and earlier.
MuleSoft announced Omni Gateway in May explicitly as a solution built for the agent control plane, governing APIs, MCP tools, LLM calls and agent traffic across MuleSoft, Kong, Apigee, AWS and Azure gateway estates. It had already put LLM governance into general availability in March. That is a Salesforce company with an installed base and a bundling motion Workato cannot answer.
Forbes reported in early July that agent gateways are becoming the control plane for enterprise AI, with a cluster of smaller entrants shipping the same week. The category is real. The question is who ends up owning it, and the honest answer is that a mid-sized independent is not the favourite.
There is one genuinely strong argument on Workato's side, and it is the neutrality claim. As Futurum notes, federation is the decisive capability, because enterprises do not migrate gateway portfolios and what platform teams actually need is consistent policy across the gateways they already run. An independent has a cleaner story here than Salesforce does — but only if the neutrality survives contact with commercial reality.
Who Is Exposed
CX, UC and contact-centre vendors. Your agent governance story is now a feature that three infrastructure vendors sell as a platform. Expect the procurement question to change from "does your agent work?" to "does your agent run under my control plane, with my identity model and my audit trail?" If the answer is a proprietary console, that is a problem.
API management incumbents. Kong, Apigee and the rest are being named as estates to be federated over rather than the control point itself.
Systems integrators. FinShield is the clearest read in the set: LTTS built a multi-agent solution in five days against a claimed two-to-three-week cycle, then published it to a marketplace. That is a repeatable channel motion, and the SIs that move first get the vertical shelf space.
Workato itself. It is competing for infrastructure status against a company that can put the same capability in a bundle a CIO has already bought.
What This Means For Buyers
You cannot have four single control planes. Somebody's is going to be the system of record and the rest are going to be things that plug into it, so decide that deliberately in the next two quarters rather than discovering it in eighteen months when four audit trails disagree.
Three questions worth more than any demo. Does it govern agents it did not build — not in the datasheet, but with a named reference doing it in production? Does the requesting user's identity propagate all the way through to the target system, or does the chain collapse into a service account somewhere in the middle? And does the audit trail reconstruct a specific action months later, to the standard a regulator would accept, not the standard a dashboard implies?
That last one is where the UK and EU angle bites. An agent drafting suspicious-activity reports is producing regulatory filings. Under the amended UK GDPR regime and FCA expectations on algorithmic accountability, the firm — not Workato, not LTTS — owns the explanation of how the output was reached. A control plane that produces a defensible record is worth paying for. One that produces telemetry is not the same thing, and vendors use the same word for both.
What Would Prove This Wrong
If the registry turns out to govern Workato-built assets well and third-party ones poorly, the neutrality claim is marketing and this is a normal product roadmap. And if enterprises simply take whichever control plane arrives bundled with Salesforce or Microsoft, the independent's better story loses to the incumbent's better invoice — which is how most infrastructure fights have ended.
Verification Note
All three Workato announcements were checked against the primary releases. FinShield's performance figures — hours to under ten minutes, 40–50% workload reduction, 20–25% cost saving — come entirely from LTTS internal benchmarking and early implementation data, with no named customer, no methodology and no independent verification; they are reported here as claims. Competitive positioning is drawn from MuleSoft's own announcements plus independent analyst coverage. Workato's "leading platform" self-description is unverified.
Sources
About the analyst: Tim Banting, 20 years in UC/CX market intelligence (Microsoft, Cisco, Omdia, GlobalData).


