As the industry gathers at Fort Mason in San Francisco for OpenAI DevDay 2026, the focus is on whether the company will pivot its core strategy. When Sam Altman takes the stage at 10am PT on September 29, he faces a market that has fundamentally shifted. The era of the general-purpose API is receding, replaced by a more aggressive paradigm: the opinionated agent.
Over the past month, major tech platforms have converged on a singular strategy: the opinionated-default pattern. Google, Apple, Meta, and Microsoft have all moved beyond providing raw model access to developers. Instead, they have shipped finished, opinionated agent products. Google launched its Search-as-agent-platform, integrating agentic booking and coding via Gemini 3.5 Flash and Antigravity for its 1 billion AI Mode users, as detailed on the official Google blog. Apple introduced Siri AI on September 14, leveraging personal context and systemwide app actions. Meta followed with Muse, a personal agent designed for multi-step task execution, while Microsoft unveiled Copilot Autopilot, a persistent, proactive agent embedded within the M365 tenant. These are not just tools; they are platforms that dictate how users interact with software.
OpenAI currently stands as the last major holdout without a dedicated, consumer-facing agent product. While the company has long been the engine of the LLM revolution, its competitors have successfully transitioned into agent platforms. DevDay 2026 serves as the critical market signal. The question is no longer whether OpenAI can build a better model, but whether it can force a platform play that rivals the integrated ecosystems of its peers.
To make this transition credible, OpenAI has begun applying a structural stick. On September 28, the company retired four legacy models: gpt-3.5-turbo-instruct, babbage-002, davinci-002, and gpt-3.5-turbo-1106. By forcing developers toward the gpt-5.6-terra replacement, as documented on the OpenAI deprecations page, OpenAI is effectively eliminating the escape hatches that allowed users to remain on older, predictable infrastructure. This move is a necessary precursor to a platform shift; by narrowing the surface area of its API, OpenAI is clearing the path for a more unified, agent-centric architecture.
Amidst this infrastructure consolidation, rumors have swirled regarding the company’s next move. TestingCatalog reported on September 26 that OpenAI may be preparing a consumer agent codenamed ‘o’, citing source code strings and a brief Pro page listing. It is important to note that this leak remains unverified and has not been confirmed by OpenAI. Whether this ‘o’ agent represents a genuine product launch or merely internal experimentation remains an open question for the developer community.
As we look toward the keynote, the focus should remain on the infrastructure. Watch for how OpenAI balances its legacy as an API-first company with the industry-wide pressure to deliver an opinionated agent. Does the company lean into the agentic future by providing a platform that competes directly with the integrated ecosystems of Google and Microsoft, or does it attempt to maintain its position as the underlying intelligence layer? The answer will define the next phase of the agent race. Pay close attention to the developer tools announced at DevDay; they will reveal whether OpenAI intends to build a platform for others to inhabit, or a product that users will inhabit themselves.