OpenAI Codex for (Almost) Everything: Release Guide

What Shipped on April 16, 2026

OpenAI framed this release around the idea that Codex should handle almost every kind of developer-adjacent work rather than narrowly coding. Underneath the marketing line, the update bundles seven distinct capability additions, each of which could have been a separate product post on its own.

The baseline reach is worth keeping in mind when reading this: about 3 million developers already use Codex every week. That user base started on the web product, expanded to the macOS desktop app on February 2, and grew again when the Windows desktop app shipped on March 4. This April update is aimed at that installed base first rather than at a brand-new user.

Who Gets It and When

The headline features are rolling out today to ChatGPT-signed Codex desktop users, but with one major caveat: background Computer Use launches on macOS, with the EU and UK arriving later. Windows users keep the rest of the desktop functionality but do not receive Computer Use at this update. Memory preview follows a similar gradual wave for Enterprise, Education, EU, and UK, which agencies running cross-border accounts should note for planning.

Background Computer Use in Detail

Background Computer Use is the headline feature, and it is deliberately different in design philosophy from the "take over the screen" computer control features competitors have shipped. Codex runs with its own cursor and its own context rather than interrupting your active work. Multiple Codex agents can run in parallel, each doing their own tasks, without stealing focus or fighting each other for the mouse.

The practical effect is that a developer can kick off a Codex task that needs to interact with a desktop app, then keep working in their IDE or browser while the agent operates in the background. For agencies used to computer-use agents that freeze the user out for the duration of a task, this is a material shift in how the tool fits into a working day.

How It Differs from Claude and Gemini Computer Use

Claude's computer-use tooling and Gemini's browser control features both lean on the model driving the user's own screen. OpenAI's design choice here is parallel background operation: several agents can hold their own cursors and work simultaneously. That unlocks workflows like running one agent through a QA pass while another fills out a CRM while a third follows up on support tickets, with the developer continuing their own work unaffected.

For a head-to-head on where each vendor lands after April 16, see our Computer Use agents 2026 matrix.

Platform and Region Constraints

• macOS: Launch platform for Computer Use.
• Windows: Desktop app shipped March 4 but does not yet receive Computer Use.
• EU and UK: Computer Use rolls out later, not at launch.
• ChatGPT sign-in required: Features are rolling out to ChatGPT-signed Codex desktop users.

In-App Browser: Frontend + Game Dev First

Codex now ships with its own hosted browser where agents render the pages, apps, and experiences they are working on. The novel part is not the browser itself but what you can do with it: comment directly on the rendered page to give the agent instructions. Instead of writing "move the CTA button up and increase the heading weight," you can annotate the button and say "move this up" and annotate the heading and say "bolder." The agent consumes those comments as instructions.

Implications for Client Work

For agency teams shipping dashboards, marketing sites, and interactive experiences, this changes the QA loop materially. Designers and PMs who previously had to translate visual feedback into prose can now annotate directly, and the agent carries the instruction straight to code. The pattern also compresses stakeholder review on early-stage builds, where the gap between "the brief said it should feel fast" and "the page feels slow here" has historically burned cycles.

If you are exploring how agents can plug into your delivery flow end-to-end, our web development practice has been integrating Codex-style agents into client sprints since the March Windows release.

gpt-image-1.5 Inside Codex Workflows

The third headline capability is image generation inside Codex workflows, powered by gpt-image-1.5. Previously, a Codex agent working on a product, frontend, or game project would stop at the point where visual assets were needed and hand off to either a designer or a separate tool. With gpt-image-1.5 available inside the workflow, an agent can generate product concepts, frontend design explorations, mockups, and game art as part of a longer task.

The creative quality is not the main point here. The main point is that image generation used to be a context switch; now it is a step. Agencies that have been stitching ChatGPT Image + Codex manually can retire that glue.

The 90+ Plugin Library

Codex now ships with a catalog of more than 90 plugins covering three previously separate categories: skills, app integrations, and MCP servers. Consolidating them into a single library matters because plugin selection used to be one of the friction points in agent rollouts. Deciding whether a capability belonged in a skill, in an integration, or in an MCP server took time that could otherwise have been spent on the actual work.

Launch Partners Worth Calling Out

• Atlassian Rovo (JIRA): Native JIRA context and actions inside Codex tasks.
• CircleCI: CI pipelines as a first-class integration, with jobs and status readable by agents.
• CodeRabbit: Review-quality context and automated PR discussion.
• GitLab Issues: Parity with the existing GitHub surface for teams not on GitHub.
• Microsoft Suite: Office, Teams, and Outlook integration for communication and docs.
• Neon by Databricks: Postgres-as-a-service integration for schema and data work.
• Remotion: Programmatic video generation inside Codex tasks.
• Render: Managed deployments without leaving Codex.
• Superpowers: Workflow automation primitives that compose with the rest of the catalog.

Developer Workflow Additions

Alongside the headline capabilities, the release bundles several developer workflow improvements that look incremental on their own but compound across a working day. These are the details that separate a demo-friendly product from something a team actually lives inside.

Summary Pane for Agent Plans

A new summary pane consolidates an agent's plans, sources, and artifacts in one view. Instead of scrolling through a long conversation to reconstruct what the agent did and why, reviewers can scan the pane to verify the plan, the sources it consulted, and the artifacts it produced. For agencies, this is a meaningful quality-control surface for anything that ships to clients.

Cross-Session Automations and Memory Preview

The third structural shift in this release, after Computer Use and plugins, is persistence. Codex can now re-enter existing conversation threads, schedule future work, and wake itself up across days or weeks. Combined with a memory preview that remembers useful context, preferences, and corrections, it crosses the line from "session-scoped assistant" into "persistent collaborator."

What Automations Actually Do

• Re-enter existing threads: Pick up a conversation from last Tuesday with full context rather than starting fresh.
• Schedule future work: Tell Codex to do something on Friday, or every Monday, and it will.
• Multi-day wake-ups: The agent wakes itself up across days or weeks to complete long-running tasks.
• Typical use cases: Landing open PRs, following up on tasks, tracking activity in Slack, Gmail, and Notion.

Practical Agency Examples

The workflows that light up here are the ones where "an assistant that forgets every Friday" was the bottleneck. Campaign retros that pull data across two weeks of Slack, dependency update sweeps that need to remember which libraries you deliberately held back, client status reports that need to track PRs and tasks across multiple repos — these all become single prompts with a schedule instead of recurring chores.

For teams wiring Codex into their CRM and operations stack, our CRM automation practice maps these patterns onto common agency operations.

Proactive Suggestions: Context-Aware Work Queues

Proactive suggestions is the piece of the release that is easiest to under-read and arguably the most behaviorally significant. Instead of waiting for a prompt, Codex scans your projects, plugins, and memory and proposes prioritized tasks. OpenAI's example was open Google Docs comments that are waiting on a reply — Codex flags them and offers to handle them.

For agencies, this is less an AI productivity feature and more a team ops feature. A project lead can start their morning by triaging Codex's suggestion queue rather than manually combing Slack, email, JIRA, and GitHub for what needs attention. The quality of the suggestions will depend heavily on which plugins are connected and how much useful memory Codex has accumulated, so early teams should expect a ramp period before the queue becomes genuinely useful.

Rollout Reality: macOS-First, EU/UK Later

Every release post glosses over the rollout constraints, and this one is no exception. The honest version is that the big-ticket items do not arrive uniformly, and agencies planning to commit to Codex for client work need to read the fine print.

For context on the Windows side of the story, our Codex Windows desktop guide covers the March 4 release and its sandboxing model in detail.

Codex vs Claude Code vs Jules After April 16

The April 16 update materially shifts the coding-agent landscape. Codex has absorbed capabilities that previously belonged to other products, specifically desktop control (from Claude's computer use) and visual iteration (from dedicated design tools). But Claude Code and Jules still hold clear ground on reasoning quality and asynchronous GitHub work respectively.

For a fuller head-to-head across benchmarks and workflows, see our Claude Code vs Codex vs Jules Q2 2026 matrix, or the companion pieces on Jules and the broader GPT-5.3 Codex release.

Agency Deployment Patterns for This Update

The temptation with a release of this size is to roll it out horizontally. The better approach is to pilot one or two high-leverage workflows that this update specifically unlocks, then expand once the team is comfortable with the new agent surface area. Four patterns are worth starting with.

For teams ready to go deeper on how to deploy coding agents at agency scale, our enterprise coding agent deployment playbook covers permissions, training, and rollout phasing in detail.

Conclusion

The April 16 Codex update earns its "for almost everything" name. Background Computer Use, an in-app browser with comment-based instructions, gpt-image-1.5 inside workflows, a unified 90+ plugin catalog, cross-session automations, memory preview, and proactive suggestions collectively turn Codex into a desktop control surface rather than a coding assistant. The 3 million-developer weekly base is now using a materially different product than the one they signed up for.

The caveats matter too. Background Computer Use is macOS-first, Windows and EU/UK users get a subset at launch, and memory preview is on a gradual rollout. Claude Code and Jules still hold clear ground on reasoning depth and asynchronous GitHub work. The right move for agencies is a one- or two-workflow pilot focused on the capabilities this release uniquely unlocks, not a full toolchain replacement.

Frequently Asked Questions