Altman Says OpenAI Will Shift to a Platform Company
Sam Altman stated that OpenAI aims to be more like a platform company rather than a single product company: providing a unified direct entry for individuals and businesses while allowing developers to build products on its intelligent capabilities through APIs.
He mentioned that ChatGPT and Codex have merged into the same entry point. Here, "merging" refers to the unification of product capabilities and interaction paths, not an immediate discontinuation of the Codex brand, independent developer tools, or APIs; Codex was originally not limited to writing code but could also perform multi-step agent tasks.
Altman's envisioned end state is "personal AGI": users would not need to switch between search, coding, research, images, voice, tool invocation, and agents, but rather have a single interface that understands the user over the long term and can access cloud computing power and external tools to complete different types of tasks. At this stage, this remains a product direction, not a delivered general artificial intelligence capability.
OpenAI plans to cover the entire cost-performance curve from low-cost to high-performance and serve more customers through a unified entry and API, rather than competing directly with its developer customers in every category of vertical software, content products, or enterprise applications. This strategy aims to reduce channel conflicts between the platform and ecosystem partners.
Altman's long-term goal is to enable 100 million new businesses and 8 billion people globally to build and use products on the OpenAI platform; this is a vision scale, not the number of customers OpenAI currently has. The company previously disclosed that its models have reached over 1 billion active users and over 2 million businesses, including ChatGPT, Codex, and ChatGPT Work.
This roadmap requires OpenAI to address issues related to models, inference power, identity, payments, memory, tool permissions, API stability, and developer distribution simultaneously. Merging ChatGPT and Codex into the same entry can lower user learning costs, but it also makes permission isolation, task controllability, error recovery, and enterprise auditing more critical product-level issues.
From a market mechanism perspective, if OpenAI successfully connects personal entry, enterprise workflows, and developer APIs, the flow of funds will shift from independent AI functional tools to model calls, agent execution, identity permissions, enterprise integration, and vertical application development. Beneficiaries will be developers who can provide industry data, exclusive workflows, and accountability based on OpenAI's capabilities; those under pressure will be products that only charge for general Q&A, basic code generation, or shallow AI packaging, lacking unique data and customer relationships.
Source: Public Information
ABAB AI Insight
OpenAI's product evolution has shifted from single-point tools to a unified intelligent entry: ChatGPT initially distributed the model in a conversational format, then added capabilities for documents, search, voice, images, projects, connectors, and agents; Codex advanced code writing, terminal execution, and multi-step tasks into the development workflow. Altman's mention of "merging" reflects the model's capabilities beginning to cross application boundaries—a single agent framework can handle writing code, researching information, creating documents, and operating tools, while product labels become an additional friction for user understanding.
The capital path has shifted from "selling a single SaaS seat" to "layered supply of intelligent capabilities." Low-cost models and free entry expand usage, while high-performance models, long-term agents, enterprise permissions, private data connections, and reliable execution increase revenue per user; APIs allow third parties to embed OpenAI capabilities into their own products. The goal of covering the entire cost-performance curve is to simultaneously capture mass scale, developer diffusion, and high-value enterprise workloads, rather than relying solely on one subscription tier.
A historical analogy is AWS evolving from infrastructure products to a development platform, and Apple forming an ecosystem with iOS, the App Store, payment, and device capabilities. AWS does not need to directly develop every type of enterprise software, nor does Apple need to operate every service application; they achieve ecosystem scale through underlying capabilities, distribution, and standard interfaces. The difference for OpenAI is that the platform does not provide stable computing resources or fixed operating systems, but rather continuously changing model capabilities, making compatibility, price stability, interpretability, and security boundaries harder to maintain than traditional API platforms.
This represents a shift in pricing power. When users centralize task entry, long-term memory, personal profiles, enterprise documents, and tool permissions into a single intelligent agent, value shifts from individual software functions to the agent coordination layer. The mechanism is: the more the model can complete tasks across applications, the less users are willing to repeatedly copy context between multiple tools; once the platform masters context, distribution, and reasoning resources, it can decide which capabilities are natively provided, which are handled by the API ecosystem, and how to price different qualities and latencies of intelligence.
ABAB News · Cognitive Laws
Functions will be absorbed by agents; entry points define platform boundaries.
Whoever controls user context controls the next layer of distribution.
The platform does not need to complete all products, only control the most expensive intelligent layer.