OpenAI Previews Specialist Dots With Microsoft Agent 365 Integration

Alongside the consumer launch of dots at DevDay 2026, OpenAI previewed the version aimed at org charts: specialist dots, agents that take on dedicated responsibilities inside an organization with their own identity and credentials for access management, IT-provisioned hardware, and deep integrations with a company's systems of record. Rollout begins as focused enterprise pilots in which OpenAI's engineering teams work directly with customers — and a partnership with Microsoft will let businesses govern these agents through Agent 365's security and management controls.

This piece covers what distinguishes a specialist dot from the personal kind, why agent identity is the pivotal enterprise design choice, what the internal-testing domains reveal, and how organizations should prepare for the pilot phase. It is written for IT, security, and operations leaders planning for a workforce that includes agents.

Specialist dots: what OpenAI previewed

A personal dot works on your behalf; a specialist dot holds a job. The company sets it up with defined responsibilities, its own identity and credentials, and access to the specific systems its tasks require. As it works, it adapts to new information and improves from the team's feedback — the employee lifecycle, restated for software. IT-provisioned hardware support extends the pattern to physical infrastructure: these agents can be furnished like staff, not just spun up like scripts.

OpenAI is deliberately not self-serving this: pilots are focused, engineering-supported engagements where each dot's duties, permitted tools, and human review process are defined jointly. The company is soliciting workloads — an explicit 'if there's work you'd like a specialist dot to take on, we'd like to hear from you.'

Why agent identity is the consequential choice

Giving agents first-class identities — rather than running them under a human's borrowed credentials or a shared service account — is the decision that makes enterprise agents governable. With its own identity, a dot's access can be provisioned and revoked precisely, its actions attributed unambiguously, and its permissions audited with the same machinery organizations already run for people. Shared-credential agents, by contrast, are invisible in exactly the ways security teams fear.

The internal testing domains OpenAI cites — procurement, invoice processing, email marketing, customer support, and commercial contracting — are all identity-sensitive: they touch money, contracts, and customer communication, where 'which actor did this' is not optional metadata but the control itself.

The Microsoft Agent 365 angle

The governance partnership is the pragmatic half of the preview. Agent 365 is Microsoft's framework for managing AI agents with enterprise controls, and integrating specialist dots means businesses can manage OpenAI's agents through Microsoft tooling they already operate. For the many enterprises whose identity and endpoint governance already runs through Microsoft, that collapses the adoption question from 'build new agent governance' to 'extend existing governance.'

The partnership is also strategically striking in a keynote that elsewhere competed hard with Microsoft's productivity turf — collaborative Pages and Slides export to PowerPoint, @ChatGPT lives inside Teams. On agents, the two are cooperating at the management layer while competing at the application layer; enterprises benefit from the détente either way.

Who should prepare, and how

The pilot profile is clear from the tested domains: high-volume, rules-plus-judgment back-office work with existing systems of record — accounts payable, vendor onboarding, support queues, contract review pipelines. Organizations with those workloads and appetite should prepare three artifacts before raising a hand: a candidate responsibility with measurable outcomes, a map of the systems and permissions it requires, and a defined human review point for the agent's outputs.

Security teams should start the policy work now regardless of pilot timing: how agent identities enter the directory, what joiner-mover-leaver looks like for a dot, and how its credential scope gets reviewed. The auto-review, Custom Rules, and monitoring safeguards OpenAI built for consumer dots carry into this context, but organizational accountability structures are the customer's to design.

Pilot metrics and Agent 365 depth: what to watch

Watch the pilots' visible outputs: which functions OpenAI showcases first, what metrics it publishes, and how quickly the program widens beyond engineering-supported engagements — that cadence signals how far specialist dots are from productization. Watch also the Agent 365 integration's depth: real policy enforcement and audit versus dashboard visibility is the difference between governance and reporting.

The framing worth adopting internally: specialist dots formalize the question every enterprise will face this decade — not whether agents join the workforce, but under whose identity, with what access, reviewed by whom. OpenAI just proposed its answers; the preview period is the time to develop yours.