OpenAI DevDay 2026 pushes agents, not just answers

Six product launches in one afternoon is not a roadmap. It is a declaration. Per Latent Space's recap of OpenAI DevDay 2026, the company used the number 1.2 billion weekly active ChatGPT users to frame a slate that included a Decisions API, an Agents API, a "Spaces" workspace product, and a Marketplace, alongside model updates (Dots, 6.1 Sol, Ultrafast). Latent Space calls it "the most confident DevDay yet." Confidence is the right word, because the announcements are not really about better answers. They are about OpenAI inserting itself into the moment where a decision gets made, and then charging for the privilege.

That is a different business than the one most B2B marketers have spent two years optimising for. The entire discipline of "getting cited in ChatGPT" assumes a static transaction: user asks, model answers, brand hopes to be named in the response. A Decisions API and an Agents API assume something else: user delegates, model acts, brand's content becomes an input to a process it cannot see and may never get named in at all. If the model books the flight, negotiates the SaaS contract, or files the compliance report, the citation moment shrinks or disappears entirely. Visibility stops being about appearing in prose and starts being about appearing in the transaction graph, as a source of record the agent trusts enough to act on without asking the user to check.

The Marketplace is the tell

Of the six launches, Marketplace deserves the most scrutiny from anyone managing brand presence, because it is the closest thing to an admission that OpenAI intends to mediate commerce, not just conversation. A marketplace implies listings, implies ranking, implies some mechanism by which one supplier's offer surfaces over another's inside an agentic flow. That is a search-engine-results-page problem wearing a different coat. Except this time the "page" is a decision the model makes on the user's behalf, with far less human scrutiny than a Google results page ever got. Financial institutions and industrial groups that spent a decade optimising for procurement portals and RFP processes now face the prospect of an AI agent doing pre-selection before a human sees a shortlist at all. Whoever gets included in that shortlid stage, and on what terms, is a question nobody in enterprise marketing has answered yet, because until DevDay 2026 nobody had to.

The scale number matters here too. 1.2 billion weekly active users is not a niche channel any more; it is close to a fifth of the world's adult population touching ChatGPT in a given week. At that scale, an Agents API that lets developers wire ChatGPT into workflows stops being a curiosity for startups and becomes infrastructure that enterprise software vendors will build against by default, the way they once built against Salesforce or SAP integrations. A UN agency or a multilateral development bank publishing guidance, standards, or funding criteria needs to ask a blunter question than "does ChatGPT cite us." The question becomes: when an agent is compiling a decision brief on regulatory compliance or grant eligibility, does it pull our framework, or a competitor's, or a Wikipedia summary of both, into its working context. That is a retrieval and structuring problem, not a content-marketing problem. It rewards institutions whose documents are already machine-legible, well-structured, and unambiguous, and punishes those whose authority lives in PDFs designed for human skimming rather than machine ingestion.