This submit is a follow-up to my earlier announcement of our protection of the agent management planes market. Analysis questionnaires for the panorama report will formally exit within the second week of April 2026.
We’re within the “dial-up web” section of the agentic period. The structure is rising quicker than the requirements wanted to make it work cleanly at enterprise scale.
In December 2025, I launched Forrester’s view of the agent management aircraft because the third useful aircraft in an enterprise agentic structure, alongside the construct aircraft and the orchestration aircraft. The thesis is that, as enterprises deploy heterogeneous brokers throughout distributors and domains, governance should sit exterior each construct and orchestration environments. A number of distributors are constructing towards this framework: The structure is sound, and vendor-agnostic management planes are inevitable. In late February, I polled 47 tech distributors, and the outcomes confirmed that:
- Seventy-nine % of taking part distributors acknowledge agent management planes as a significant and distinct product class.
- Ninety-two % have assigned a named product supervisor or group to agent governance or management aircraft performance.
- Forty % report lively RFPs or buyer shopping for motions that explicitly request a management aircraft or equal.
That stated, enterprises at present battle to implement the management aircraft as a conveyable, vendor-agnostic governance layer as a result of the requirements stack beneath it’s incomplete.
Requirements Lag Behind Architectural Finest Follow
Forrester’s three-plane mannequin decomposes the enterprise agentic “stack” into three distinct planes: “construct,” “orchestrate,” and “management.” The important thing hurdle to realizing a clear useful stack alongside this framework is that the connective tissue between the planes, requirements, and protocols that permit governance choices in a single aircraft to propagate reliably into one other stays underdeveloped. Three classes of requirements gaps create three distinct boundaries to creating a constant management aircraft operational at enterprise scale.
Barrier 1: Instrumentation Requirements Are Incomplete
A management aircraft can’t govern what it may well’t observe. The first instrumentation customary for agentic AI telemetry is OpenTelemetry’s genAI semantic conventions, which now cowl mannequin operations, agent creation and invocation, instrument execution spans, analysis occasions, and multimodal content material. Latest releases added agent model attributes, retrieval span help, and cache token monitoring. Datadog introduced native help for genAI semantic conventions at model 1.37 and above in late 2025, permitting groups to instrument as soon as with OpenTelemetry and export genAI spans via present collector pipelines. The momentum is constructing, however the conventions themselves stay experimental. The OpenTelemetry mission hasn’t but printed a steady model of the genAI semantic conventions, which suggests each adopter builds on a shifting goal. Extra importantly, the present conventions tackle operational telemetry (spans, metrics, and traces for mannequin calls and power executions) however don’t but cowl the total governance floor a management aircraft requires. Ability-level identification propagation, value attribution traced to enterprise worth streams, and cross-orchestrator span correlation sit exterior the present specification’s scope. Instrumentation requirements can inform you what occurred inside an agent’s execution, however they’ll’t but inform you who the agent was in governance phrases, what enterprise coverage utilized to it, or how interventions ought to propagate throughout platforms.
A parallel effort addresses the monetary telemetry hole. The FinOps Basis’s FOCUS specification (FinOps Open Price & Utilization Specification) normalizes billing knowledge throughout cloud, SaaS, AI workloads, and knowledge middle spend. The State of FinOps 2026 Report discovered that 98% of respondents now handle AI spend, up from 63% in 2025, and that AI value administration ranks as the highest forward-looking precedence for FinOps groups globally. FOCUS addresses a distinct dimension of the management aircraft downside than OpenTelemetry does: monetary telemetry quite than operational telemetry. Each should converge for a management aircraft to perform as designed. And neither, by itself, solves the deeper downside: The agent’s governance identification doesn’t but journey with it.
Barrier 2: Agent Id And Coverage Propagation Lack Moveable Requirements
That is probably the most consequential hole on which the opposite two rely and the one which connects all three boundaries. When a developer wires an agent to a selected mannequin, grants it entry to a set of instruments, and deploys it right into a runtime atmosphere, that agent carries a composite identification: mannequin bindings, instrument bindings, permission scopes, value ceilings, and behavioral constraints. For the management aircraft to control that agent at runtime, the identification should journey with the agent from construct via deployment into manufacturing in a standardized format. No such customary exists on the stage of maturity that enterprises require. With out a moveable agent identification descriptor that crosses all three planes, instrumentation (barrier 1) can’t totally describe the agent and integration schemas (barrier 3) don’t have any identification anchor to reference.
The protocol panorama displays the issue. Mannequin Context Protocol (MCP) handles agent-to-tool connectivity and has achieved extraordinary adoption, with a number of million month-to-month SDK downloads and governance underneath the Linux Basis’s Agentic AI Basis. MCP model 2.1 launched server identification and enhanced safety features. Google’s A2A (Agent2Agent) Protocol handles multiagent coordination with agent playing cards that describe agent capabilities. IBM’s BeeAI Agent Communication Protocol makes use of agent manifests for the same objective. Microsoft’s Entra Agent Registry builds a manufacturing implementation of agent manifest-based discovery inside a proprietary identification infrastructure. Every protocol addresses an actual want, and the fragmentation displays genuinely totally different design scopes quite than competing makes an attempt on the identical downside. However none of them clear up the moveable agent identification downside throughout all three planes as a result of none of them had been designed to.
NIST acknowledged this hole straight. In February 2026, NIST’s Heart for AI Requirements and Innovation launched the AI Agent Requirements Initiative, organized round three pillars: facilitating industry-led agent requirements, fostering open-source protocol growth, and advancing analysis in AI agent safety and identification. NIST’s Nationwide Cybersecurity Heart of Excellence launched an idea paper titled “Accelerating the Adoption of Software program and AI Agent Id and Authorization,” exploring how present identification and entry administration requirements can apply to AI brokers working throughout enterprise infrastructure. Public feedback shut on April 2, 2026. This initiative represents the primary formal institutional effort to coordinate identification governance for autonomous AI methods on the federal stage.
On the decentralized facet, the Agent Community Protocol makes use of World Extensive Internet Consortium (W3C) decentralized identifiers (DIDs) for cryptographic agent identification, and the W3C AI Agent Protocol Neighborhood Group targets official net requirements for agent communication by 2026–2027. DIDs signify the closest factor to a conveyable identification primitive for brokers, however adoption stays early-stage and concentrated in interorganizational situations quite than intra-enterprise governance.
Each main vendor and requirements physique sees the identical want. Each one builds a barely totally different reply. Till a conveyable agent identification descriptor exists that may journey throughout construct, orchestrate, and management planes, enterprises will hand-build the propagation logic for each integration. That limits the management aircraft to platform-specific implementations quite than the vendor-agnostic governance layer the structure requires. It additionally signifies that the third barrier, the absence of cross-plane integration schemas, has shaky foundations on which to construct.
Barrier 3: Cross-Aircraft Governance Schemas Don’t Exist
Even when OpenTelemetry stabilizes its genAI conventions and the {industry} converges on a conveyable agent identification customary, a 3rd layer of requirements stays absent: the schemas that outline how the construct, orchestrate, and management planes change governance-relevant data about agent state, coverage, and lifecycle.
Contemplate what these schemas would want to specific. When a management aircraft points a coverage change (e.g., revokes an agent’s entry to a instrument, lowers its value ceiling, or requires human approval for a category of actions), that change should propagate into the orchestration layer as an enforceable constraint on workflow execution and into the construct layer as a configuration replace or deployment gate. That requires a standardized coverage propagation object: a machine-readable directive that any orchestration platform or CI/CD pipeline can eat with out bespoke integration. When an orchestration platform detects that an agent’s conduct has drifted exterior its anticipated efficiency envelope, it should emit a standardized lifecycle occasion (not only a telemetry span however a governance-grade sign) that the management aircraft can act on: droop, reroute, throttle, and/or escalate. When a construct instrument publishes a brand new agent or updates an present one, it should produce a functionality manifest, a declared contract describing the agent’s mannequin bindings, instrument entry, permission scope, and behavioral constraints — all in a format that the management aircraft can ingest and implement at runtime.
What This Means For Enterprise Leaders
None of those gaps invalidate the case for an agent management aircraft. Quite the opposite, the gaps validate the structure by figuring out the boundaries the place requirements want to come back into existence. Enterprises nonetheless want the conceptual separation between construct, orchestrate, and management to make sound long-term platform choices — even when the connective tissue between planes stays hand-built for now.
Every barrier carries a selected implication:
- Towards barrier 1: Instrument early. Undertake OpenTelemetry’s genAI semantic conventions now, even of their experimental state, and align your FinOps observe with FOCUS. The price of retrofitting instrumentation later far exceeds the price of adopting evolving conventions at present. Constructing the observability basis now offers the management aircraft one thing to control when it matures.
- Towards barrier 2: Monitor the identification requirements panorama actively. NIST’s AI Agent Requirements Initiative, the Agentic AI Basis’s stewardship of MCP, and the W3C’s AI Agent Protocol Neighborhood Group signify the three most consequential efforts shaping how agent identification and authorization will work throughout enterprise boundaries. The choices these our bodies make over the subsequent 12–18 months will decide which architectural bets repay and which go away you locked right into a single vendor’s identification mannequin.
- Towards barrier 3: Design for aircraft separation now. Enterprises that conflate build-time governance, orchestration-time governance, and runtime governance right into a single undifferentiated “agent administration” perform will face costly architectural refactoring when cross-plane integration schemas emerge and the market consolidates across the three-plane mannequin.
The person requirements are actual and progressing. The mixing between them is no person’s job but — that can change quickly. Architect for this separation of planes at present so that you simply’re properly positioned to undertake the connective tissue when it arrives.
