Brockman: doubling down on text

Brockman just said the quiet part out loud: OpenAI is betting that the future of artificial intelligence still runs through text, not through flashy demos alone. In a recent interview, OpenAI president Greg Brockman said the company is “doubling down on text models,” sees “line of sight” to much better systems this year, and wants tools like Codex to sit closer to the center of the product. (bigtechnology.com) That sounds almost backward if you only watch the headlines. The public conversation around artificial intelligence has been pulled toward video generators, image models, voice assistants, and humanoid robots, but Brockman’s framing says OpenAI still thinks language is the control layer that ties everything else together. (bigtechnology.com) Text models are not just chatbots in this view. They are systems that take instructions, hold context across many steps, call tools, inspect outputs, revise plans, and keep going until a task is finished, which is closer to how a project manager works than how a search box works. (bigtechnology.com) That matters because most real work inside a company is still expressed in language. Product requirements are written in documents, software bugs are described in tickets, legal reviews happen in comments, customer support runs through messages, and strategy gets passed around in memos and spreadsheets with text wrapped around the numbers. (bigtechnology.com) If you can build a model that reads, writes, reasons, and acts inside those workflows, you do not need to replace every application at once. You can sit above the existing stack like an operating layer, taking plain-English requests and turning them into sequences of actions across coding tools, browsers, documents, and internal systems. (bigtechnology.com) That is where Codex enters the story. OpenAI now describes Codex as a coding agent for software development, available across ChatGPT Plus, Pro, Business, Education, and Enterprise plans, and designed to write code, explain codebases, and help with software tasks end to end. (developers.openai.com) OpenAI’s own product language around Codex is revealing. The company says Codex can handle everything from writing code to understanding unfamiliar repositories, and the ChatGPT-facing Codex page describes it as a system that can complete routine pull requests, refactors, and migrations, not just suggest snippets in a text box. (developers.openai.com) (chatgpt.com) Once a coding agent moves from “answer my question” to “finish this task,” the product logic changes. The model is no longer just a smarter autocomplete tool; it starts to look like a worker that lives inside a broader workspace, where chat, browsing, coding, and execution belong in one place instead of four separate windows. (developers.openai.com) (chatgpt.com) That lines up with OpenAI’s reported super-app plan. CNBC reported on March 19, 2026, that OpenAI plans to combine its web browser, ChatGPT app, and Codex coding app into a single desktop application, with the goal of reducing fragmentation and streamlining the user experience. (cnbc.com) In that setup, text becomes the universal interface. You ask for a market analysis, the system opens sources, compares numbers, drafts a memo, writes a slide outline, and maybe generates a script for a dashboard, all from one thread of instructions instead of a chain of copied-and-pasted prompts. (bigtechnology.com) (cnbc.com) Brockman also tied this product direction to model progress. In coverage of the same interview, he said OpenAI’s GPT reasoning models have “line of sight” to artificial general intelligence, which suggests the company sees better reasoning in language as the path to broader capability, not as a side feature attached to other media formats. (the-decoder.com) (bigtechnology.com) There is also a timing signal here. Brockman said OpenAI sees a path to “much better systems this year,” and OpenAI’s own news page shows the company is actively expanding Codex as a product, including a post on April 2, 2026 announcing pay-as-you-go pricing for teams, which usually means a tool is moving from experiment toward wider operational use. (bigtechnology.com) (openai.com) Put all of that together and the message is pretty clear. OpenAI is not backing away from multimodal artificial intelligence, but its leadership is publicly signaling that text-centric reasoning, paired with tools that can actually do work, is still the foundation it wants to build on. (bigtechnology.com) (cnbc.com) The simplest way to read Brockman’s comments is this: OpenAI thinks the winning artificial intelligence product may look less like a gallery of separate model tricks and more like one place where you type a request and the system carries the job through. If that is right, “doubling down on text” is not a retreat to the old chatbox era; it is a bet that language will remain the command center for everything that comes next. (bigtechnology.com) (developers.openai.com)