Microsoft Just Wrote the Agentic AI Playbook. Here Is What It Leaves Out.
Dispatches from the Agentic Frontier is a new regular intelligence briefing for leaders in knowledge-intensive sectors. Each dispatch translates evidence from the frontier of agentic AI — practitioner experience, investor signals, strategy research, market events — into what it means for enterprise leaders building for competitive advantage. Some dispatches are field reports from practitioners who are 12–18 months ahead. Others synthesise emerging research. All are filtered through the Intelligence Capital framework developed in The AI Your Competitors Can’t Buy.
Today we review Microsoft’s recently published enterprise agentic AI playbook, built from three years of their own company-wide deployment — a “Customer Zero” retrospective from the vendor whose infrastructure most of the enterprise world runs on.
It is genuinely useful. But it describes a destination that contains no mechanism for compounding advantage. Every enterprise running on Azure should study it and realise that durable advantage starts where the playbook stops.
Introduction
On 16 April 2026, Microsoft Digital — the company's IT organisation — published Becoming a Frontier Firm: A guide for deploying AI agents based on our experience at Microsoft. It is 10,000 words of governance frameworks, maturity models, adoption methodology, and measurement strategy, drawn from Microsoft's own internal deployment as "Customer Zero" for agentic AI. Microsoft accompanied the guide with the launch of Agent 365, its new control plane for managing agents across the enterprise.
This is not a marketing document. It is the most detailed account any major technology vendor has published of how to deploy, govern, adopt, support, and measure AI agents at enterprise scale. The governance matrix alone — distinguishing no-code agents requiring no review, low-code agents requiring environment controls, and pro-code agents requiring full security, privacy, and responsible AI reviews — gives CIOs and CTOs a ready-made classification framework they can adapt immediately.
The maturity model, the adoption methodology, the Agent 365 registry — all of it is genuinely useful. But none of it compounds.
What Microsoft Got Right
Three moves in the guide are available to any organisation, regardless of size or sector.
The governance matrix differentiates agent oversight by complexity and risk, not by a blanket policy applied uniformly. An employee building a knowledge-only agent from pre-approved SharePoint sources requires no proactive review. A professional developer building an enterprise workflow agentic system with custom connectors and write access to external systems requires cross-disciplinary review across security, privacy, accessibility, and responsible AI. This risk-proportionate approach is the right model. Most organisations have not yet built it.
The five-stage maturity model — from awareness through pilots, operationalisation, enterprise adoption, and agentic transformation — provides a sequenced roadmap that stages investment against executive readiness, data readiness, and workforce readiness. The stages are generic, but the structure is sound: it prevents organisations from scaling agents before they have the governance foundation to do so safely.
Agent 365, now in early access, provides a registry, visualisation, interoperability, and security layer across an organisation's entire agent estate. For enterprises struggling with agent sprawl — duplicate agents, unclear ownership, no way to identify the authoritative source — this is a significant operational advance.
These are the floor of serious enterprise agentic AI. Every Azure-using organisation should adopt them.
The Frontier Firm as Microsoft Defines It
Microsoft's destination is what it calls a "Frontier Firm" — an organisation that is "AI-operated but human-led." The guide describes three interaction patterns: humans working with an AI assistant, human-agent teams, and humans leading teams of agents that perform core work with relative autonomy. Five stages of maturity culminate in "transforming your business with agentic AI." Six dimensions of value — strategy, analytics, accelerators, change management, governance, and lifecycle — structure how the transformation is measured and managed.
This framework was written by Microsoft Digital, the company's IT organisation. That authorship matters.
In a previous dispatch, I identified one of Jamie Dimon's most consequential strategic moves at JPMorgan: taking AI out of IT and placing it on the Operating Committee under a business-side veteran. The logic was precise: AI transformation is a business redesign challenge, not a technology deployment programme, and IT organisations think in governance, tooling, and lifecycle management, not in institutional reasoning or permanently-owned competitive assets.
The Microsoft guide is the product of IT staying in IT's lane. Every chapter reflects it. Governance, implementation, adoption, support, measurement — these are the concerns of an IT organisation deploying technology well. They are not the concerns of a business leader asking what the organisation will own when the deployment is complete.
For CTOs, the framework is correct within its scope. For CEOs and COOs who read it as the enterprise AI strategy, it is a map of the infrastructure layer mistaken for the territory.
The Missing Word
The word "deliberation" does not appear in the 10,000-word guide. The word "memory" appears once, in passing. "Intelligence" appears frequently, but always as "machine intelligence" — compute capacity — never as an organisationally-owned asset that accumulates with use.
The absence is not an oversight. The framework is internally coherent without these concepts. Microsoft's five stages describe how to deploy agents well, govern them safely, adopt them broadly, support them effectively, and measure their impact precisely. An enterprise that follows the framework to its natural conclusion arrives at a governed, measured, adopted agentic estate that produces no compounding advantage.
Microsoft measures impact in six dimensions. All six are efficiency metrics — adoption, usage, time savings, cost reduction, quality improvements. None measures what the organisation has accumulated that competitors cannot later acquire by purchasing the same technology. The "Frontier Firm" destination, as Microsoft defines it, is a firm that has automated more and governed better. It is not a firm that has built an asset its competitors cannot replicate.
This is the Parity Problem at the infrastructure layer. PwC's recent AI performance study, surveying 1,217 companies across 25 sectors, found that 20% of companies capture 74% of AI-driven returns — concentration driven by execution discipline, not by accumulated institutional advantage. The top quintile achieves 7.2 times the AI-driven performance of the rest. That gap reflects precisely the kind of governance, adoption, and measurement discipline Microsoft describes. It does not reflect Intelligence Capital.
What the Framework Inadvertently Confirms
Microsoft devotes significant attention to agent sprawl — employees creating redundant agents, no lifecycle management, no way to identify which agent is the authoritative source. Agent 365 is designed to solve this problem through registry, observability, and governance. The solution is sound for an IT management problem.
But sprawl is a symptom of a deeper architectural absence. In an Intelligence Capital-oriented architecture, agents are not personal productivity tools that proliferate according to individual preference. They are components of an organisational intelligence system with persistent memory, clear authority hierarchies, and deliberation capture as a first-class output. The sprawl problem Microsoft describes does not arise when the architecture is designed for reasoning accumulation rather than individual task acceleration.
Agent 365's interoperability promise is real but structurally constrained. The guide states that Agent 365 is "open to any Microsoft-built or partner ecosystem" — meaning interoperability within Microsoft's identity and governance infrastructure. Cross-boundary agent trust still depends on Microsoft's own identity stack. Organisations that build their entire agent governance layer within the Microsoft environment will accumulate operational history that is portable within Microsoft's infrastructure and not beyond it.
This is not a criticism of Microsoft's product quality. It is an observation about what the product is designed for. Agent 365 is a management and observability layer. It tells you what agents exist, how they perform, and whether they comply. It does not connect the intelligence generated by one agent to the decisions made by another. It does not capture deliberation. It does not create cross-functional feedback loops where insight generated in one domain enriches every other. The distinction matters: Microsoft is building the monitoring infrastructure, not the Coordination Layer.
The Counterfactual
Consider two commercial insurers, both running on Azure, both following Microsoft's guide.
Insurer A deploys Agent 365, registers its AI agents supporting claims, governs them through Microsoft's tiered review process, tracks adoption and time savings. Its claims agentic system logs activity for audit. The system operates well. After ten thousand claims, Insurer A can report efficiency gains, adoption rates, and quality improvements.
Insurer B deploys the same governance framework — but architects its claims agentic system with deliberation capture as a first-class output. Every adjuster override, every coverage interpretation, every reserve decision is recorded with structured reasoning: why this decision was made, what was weighed, what precedent it reflects. After ten thousand claims, Insurer B owns institutional memory that no competitor can purchase — not from Microsoft, not from any vendor. The ten-thousand-and-first claim is qualitatively better than the first, not because the model improved, but because the organisation's reasoning deepened.
The technology is identical. The governance is identical. The architecture is not. One insurer has deployed well. The other has built an asset.
The April 2026 Consensus
Last week, I examined what McKinsey, BCG, and Bain are missing about agentic AI — the consultancy consensus that correctly identifies the urgency and depth of AI transformation but stops short of identifying the mechanism that makes advantage permanent rather than temporary. Microsoft's guide completes the picture from the other side: the infrastructure vendor defining the deployment destination, with the same gap.
The advisory firms tell you how to think about transformation. The infrastructure vendor tells you what to build. Neither tells you what to own.
The convergence is worth stating plainly. The April 2026 consensus on "what good looks like" in enterprise agentic AI — authored independently by the world's most influential strategy firms and the world's dominant enterprise AI vendor — is remarkably coherent on governance, adoption, and measurement. It is also remarkably silent on the one capability that produces durable advantage: the deliberative accumulation of institutional reasoning that compounds with every decision, persists independently of any individual, and cannot be replicated by purchasing the same technology later.
That capability is Intelligence Capital. And it sits above the ceiling that every published framework defines as the destination.
The Choice
Microsoft has handed every enterprise the most substantial operational playbook for agentic AI any platform vendor has produced — built from lived experience, not theory. It’s well worth following. The governance, the maturity staging, the adoption discipline, the Agent 365 observability — all of it is the right foundation.
Then ask the question Microsoft's guide does not: what will your organisation own when the deployment is complete?
If the answer is a governed, well-adopted, efficiently measured agent estate — you have followed the playbook. You are a Frontier Firm. But you will discover, within eighteen months, that your competitors who followed the same playbook have arrived at the same destination. The Parity Problem is structural, not accidental.
If the answer is a Coordination Layer that captures deliberation as a permanently-owned institutional asset — connecting the intelligence generated in one domain to the decisions made in every other, compounding with every case, and owned by the organisation regardless of which platform generated it — you have built the layer Microsoft did not write.
The playbook is the floor. The Coordination Layer is the architecture above it. Technology can be purchased at any point. Time cannot. The organisations that begin capturing deliberation now will compound an advantage that late movers cannot close — because they will lack the reasoning history, the institutional memory, and the cross-functional intelligence that only accumulate through elapsed time.
Microsoft's map ends where the strategic question begins. Where the map ends is exactly where the territory gets interesting…
Simon Torrance is CEO of AI Risk, an agentic AI strategy and innovation advisory firm. AI Risk's Intelligence Capital framework builds on the foundational work of Professor David Shrier (Imperial College London). For more on the Agentic AI Maturity Model, the Parity Problem, and the Coordination Layer, see The AI Your Competitors Can't Buy.
