The editorial argues that broadening Codex from a dedicated software engineering agent with SWE-bench claims to an 'almost everything' agent is the language companies use when they have an unresolved positioning problem. It reads the move as OpenAI ceding the specialist coding-agent narrative to Claude Code, Cursor, Devin, and Jules rather than competing head-on.
The thread's highest-voted responses converged on reading the marketing shift as a retreat from a year ago when Codex was pitched as a dedicated software engineering agent. Commenters interpret the vague 'everything' framing as a tell that OpenAI is struggling to differentiate in the now-crowded agentic coding market.
OpenAI positions Codex as the default entry point for any agentic task an OpenAI customer wants to run, emphasizing autonomous multi-step work in a sandbox, tight integration with ChatGPT projects and the desktop app, and background tasks that return PRs, spreadsheets, or research briefs. Bundling into existing Plus, Pro, and Business tiers signals they want Codex to be the ambient agent layer across the product, not a niche coding SKU.
The editorial observes that by April 2026 Claude Code owns terminal-resident workflows for senior backend/infra engineers, Cursor owns the IDE for frontend and full-stack teams, Devin occupies async ticket-to-PR work at larger orgs, and Jules has captured Gemini-native shops. In that landscape, a generalist Codex is entering a field where incumbents already have segment-specific mindshare, making breadth a harder sell than depth.
OpenAI published 'Codex for Almost Everything,' positioning its Codex product line as a general-purpose agent that spans software engineering, research, and day-to-day task delegation. The post hit 475 points on Hacker News within hours. This is the third material repositioning of the Codex name: the original 2021 code-completion model, the 2025 cloud-based coding agent relaunch, and now a 2026 push to make Codex the default entry point for anything an OpenAI customer wants an agent to do.
The new framing emphasizes three capabilities: autonomous multi-step work inside a sandboxed environment, tighter integration with ChatGPT projects and the desktop app, and first-class support for running background tasks that return with a PR, a spreadsheet, or a research brief. Pricing and rate limits remain tied to existing ChatGPT Plus, Pro, and Business tiers rather than a standalone SKU, which is a deliberate bundling move.
The HN thread's top-voted replies converged on a single observation: the marketing is a retreat, not an advance. A year ago OpenAI was selling Codex as a dedicated software engineering agent with benchmark claims against SWE-bench. Today it is selling 'almost everything,' which is the language companies use when they have a good general-purpose product and an unresolved positioning problem.
The agentic coding market in April 2026 is not the open field it was eighteen months ago. Anthropic's Claude Code has become the default terminal-resident agent for a large slice of senior engineers, particularly in the backend and infra segments where its long-context editing and tool-use discipline matter. Cursor still owns the IDE experience for frontend and full-stack teams. Cognition's Devin found a niche in async ticket-to-PR workflows at larger orgs. Google's Jules shipped, stabilized, and pulled a respectable share of Gemini-native shops. Into that crowded field, OpenAI is now pitching Codex as the agent that does coding *and* everything else — the classic platform move when the vertical fight is going sideways.
The 'almost' is the most honest word in the announcement. What Codex still does not do well, based on the community reports filtering through the thread and early hands-on posts: sustained multi-repo refactors where context has to be rebuilt across sessions, anything requiring a truly local execution environment (Codex's cloud sandbox is a feature for safety and a limitation for speed), and work that depends on enterprise-internal tooling not yet exposed via MCP or OpenAI's function interface. A recurring complaint is latency: the sandboxed execution model adds round-trips that Claude Code, running directly in your terminal with your actual file system, simply does not pay.
The comparison that matters for anyone evaluating a switch is not benchmark scores — those converge quickly and the gaps close within a model generation. It is workflow friction. Claude Code's edit-apply-test loop sits inside the developer's existing shell. Cursor's inline diff flow sits inside the existing IDE. Codex's strongest experience still pulls the developer into ChatGPT's surface, which is where OpenAI wants the center of gravity but where experienced practitioners are often least comfortable working. For teams already shipping with Claude Code or Cursor, the switching cost isn't the model — it's retraining muscle memory on a different orchestration surface.
There is a second, quieter story here. OpenAI's willingness to collapse Codex, research agents, and general task delegation into one product is a bet that the distinction between 'coding agent' and 'agent' will dissolve over the next two years. That bet is probably correct. It is also the same bet Anthropic is implicitly making by turning Claude Code into the orchestration layer for arbitrary tooling via MCP. The disagreement is about surface: OpenAI wants you inside ChatGPT, Anthropic wants you inside the terminal and your IDE. Neither vendor benefits from a neutral layer winning — which is precisely why the neutral layer (MCP, increasingly OpenCode) deserves more of your attention than either announcement.
If you already have a working agent workflow, do not switch on the basis of this announcement. The benchmark delta between Codex, Claude Sonnet 4.5, and the current Gemini coding tier is within noise for most real engineering work. The variance you feel day-to-day comes from context management, tool availability, and the specific shape of your repo, not raw model quality. Changing agents costs roughly two weeks of productivity per developer, and Codex's 'almost everything' pitch does not clear that bar for teams with an existing flow.
If you are greenfield, the decision is more interesting. Codex's bundling into ChatGPT Business makes it the cheapest serious option for shops already paying for OpenAI seats — you are not buying a new product, you are turning on a feature. The gotcha is execution model: if your work requires heavy local tooling (CUDA kernels, embedded toolchains, anything with a long-running local daemon), the cloud sandbox will bite. If your work is mostly cloud-native API stitching, data pipeline authoring, or standard web application work, the sandbox is fine and the bundling wins.
For teams running evals internally, the actionable move this week is to rerun your harness against Codex's new agent mode with your actual repo and your actual failure cases — not SWE-bench, which all three major vendors have now implicitly trained against. Track three numbers: mean time from prompt to mergeable PR, rate of PRs that require human rewrite rather than review, and token spend per successful task. The last one is where surprises tend to live; Codex's sandboxed retries can run up bills fast on tasks it cannot complete.
The shape of the agentic coding market by the end of 2026 looks like two or three integrated platforms (OpenAI, Anthropic, Google) and one neutral protocol layer (MCP plus whatever OpenCode becomes), with the terminal and the IDE as the two contested surfaces. OpenAI's 'almost everything' framing is a signal that the product-category fight is over and the platform fight is beginning. The vendors that win the next phase will be the ones that make the agent invisible — it lives where you already work, in the surface you already use, against the codebase you already have. Everyone currently pitching a new UI is losing ground. Codex's bet on ChatGPT-as-surface is the riskiest version of that pitch. Whether it pays off will depend less on model quality and more on whether OpenAI can convince senior developers to move their center of gravity out of the terminal — which, on current evidence, is not a bet worth taking.
There seems a fair enthusiasm in the UI of these to hide code from coders. Like the prompt interaction is the true source and the actual code is some sort of annoying intermediate runtime inconvenience to cover up. I get that productivity can be improved with a lot of this for non developers, just n
Lots of scepticism here, but I think this may really take off. After 25 years of heavy CLI use, lately I've found myself using codex (in terminal) for terminal tasks I've previously done using CLI commands.If someone manages to make a robust GUI version of this for normies, people will lap
I've been using the Codex app for a while (a few months) for a few types of coding projects, and then slowly using it for random organizational/productivity things with local folders on my Mac. Most of that has been successful and very satisfying, however...Codex is still far from ready fo
Tried it out. It's a far more reasonable UI than Claude Desktop at this moment. Anthropic has to catch up and finally properly merge the three tabs they have.The killer feature of any of these assistants, if you're a manager, is asking to review your email, Slack, Notion, etc several times
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
My current expectation is that the Cowork/Codex set of "professional agents" for non-technical users will be one of the most important and fastest growing product categories of all time, so far.i.e. agents for knowledge workers who are not software engineersA few thoughts and question