Specialist Dots: The Early Pilots Nobody's Talking About Yet

Most of the coverage of OpenAI's September 29 Dots launch focused on the consumer-facing version: a personal agent with its own cloud computer, reachable from Slack and Teams, that learns your habits over time. Tucked into that same announcement, with far less fanfare, was a program that's arguably more consequential for how AI agents actually get deployed inside real companies — Specialist Dots.

What's actually confirmed

Per VentureBeat's coverage of the launch, OpenAI describes Specialist Dots as persistent agents with organizational identities, credentials, and defined business responsibilities — provisioned through a company's existing systems the way a new employee would be. The company has tested agents in five specific functions: procurement, invoice processing, email marketing, customer support, and commercial contracting. That's a notably concrete list, more specific than most of the general-purpose "dots can help with X" language elsewhere in the launch materials, and it suggests these pilots have been running for some time before the public announcement rather than being conceived for the DevDay stage.

Each of those five functions shares a common trait: they involve an agent acting with real organizational authority on a company's behalf, often touching money, legal terms, or customer-facing communication. That's a categorically different risk profile than a personal dot drafting a social media post, and it's why OpenAI paired this program with Agent 365 governance through Microsoft, treating each Specialist Dot as what the company calls a "machine identity" requiring onboarding, credentials, access policies, monitoring, and offboarding. Agent 365 and the Enterprise Path Microsoft Is Building for Dots covers that governance layer in full.

What's genuinely undisclosed

Commercial pricing for Specialist Dots has not been published — a distinct gap from the general "additional dots beyond your first free one" pricing that's also undisclosed. This is worth separating out clearly, because a Specialist Dot provisioned with organizational credentials for, say, invoice processing is a fundamentally different commercial product than a second personal dot for an individual Pro subscriber, and enterprise buyers evaluating this seriously will want to know it's priced and packaged separately before assuming anything about cost.

OpenAI also hasn't disclosed which companies are running these pilots, how long they've been running, or what results they've produced. The five named functions read like a reasonable, sensible starting list for agentic automation — all of them are process-heavy, rules-governed, and already partially automated in most organizations — but "OpenAI is piloting this" and "this works reliably in production" are different claims, and only the second one would justify broad enterprise adoption. Treat the list of functions as OpenAI's stated intent, not as evidence the pilots have matured past early testing.

A specific list of five business functions is a more credible signal of real deployment than any amount of consumer-facing demo footage — but a list of pilots isn't a track record, and OpenAI hasn't offered one yet.

Why procurement and contracting are the interesting cases

Customer support and email marketing are relatively low-risk domains for early agent deployment — mistakes are visible and correctable, and most companies already run some automation there. Procurement and commercial contracting are a different order of stakes: an agent with a company identity approving a purchase order or negotiating contract terms is making decisions with real financial and legal consequences, decisions that traditionally require a human signature specifically because the cost of getting them wrong is high and hard to reverse.

That's precisely the kind of high-consequence action Auto-Review — the mechanism that checks a dot's more consequential actions against a user's instructions and built-in safety requirements before execution — is meant to gate. Whether Auto-Review, as currently built, is robust enough to be trusted with procurement or contracting decisions at the volume a real enterprise would run them is an open and important question that OpenAI's announcement doesn't answer. Auto-Review: The Safety Net Between a Dot and a Consequential Action covers what is and isn't known about that check, and it's worth reading with these two specific pilot functions in mind — this is exactly the use case the mechanism will be judged on.

It's also worth noting how differently this reads next to the other two products in this launch wave. Grok Bot's business rollout leans on the user's own credentials signing into existing tools directly, with no equivalent formal governance framework confirmed in the sources reviewed for this series — a much less structured approach to exactly this kind of high-stakes deployment. Meta's Muse, by contrast, keeps its agent strictly out of passwords and payment methods via the Sentinel approval layer, a design choice that would make something like autonomous procurement approval a much harder fit for Muse's current architecture. Specialist Dots, whatever its actual maturity, is at minimum the most structurally serious attempt among the three to build governance for exactly this category of high-stakes agentic work — which is also why its limits, once real customers start hitting them, will be worth watching closely.

What to watch for as these pilots go public

If Specialist Dots moves from quiet pilot to general availability, the details worth watching for are less about the headline capability and more about the fine print: what audit trail a company gets when a Specialist Dot approves a purchase order or modifies contract terms; whether there's a clean mechanism to revoke a Specialist Dot's credentials immediately if something goes wrong, the way a company would disable a compromised employee account; and whether OpenAI publishes any track record — error rates, incident reports, customer testimony — beyond the initial list of five functions. None of that exists in public form yet. Given how much of this series has had to flag pricing, safety documentation, or reliability data as simply undisclosed across all three vendors, Specialist Dots reaching general availability with a genuinely transparent account of its own limitations would be a meaningful signal that OpenAI is treating enterprise trust as something to earn with evidence, not just architecture.

Part 19 of 30 in the ChatGPT Dots series. Previously: Agent 365 and the Enterprise Path Microsoft Is Building for Dots. Next: What a Dot Cannot Do Yet: The Honest Limits List. Browse more builder insights or explore AI skills at aiskill.market.