H HiddenPro AI

How to Turn Screenshots, App Windows, and Half-Finished Work Into a Codex Workflow That Keeps Moving

May 25, 2026

OpenAI’s May 14, 2026 and May 21, 2026 updates changed what it means to “use Codex.” The workflow is no longer just about typing a request into a blank chat box and hoping the model reconstructs your context. With the Codex app in the ChatGPT mobile app, appshots, goal mode, and newer support for longer-running work, you can hand off a real screen, a real task, and a clearer finish line instead of re-explaining everything from scratch.

That matters because most knowledge work does not start cleanly. It starts with a slide deck in progress, a browser tab full of references, a dashboard that needs interpretation, or notes that are only half-finished. The Codex appshots workflow is built for that reality: capture the context that already exists, define what “done” should look like, and let Codex keep moving while you step away, then review the result with less back-and-forth.

Why the new Codex workflow matters now

The May 14, 2026 release introduced Codex in the ChatGPT mobile app in preview, which made it easier to stay connected to work that begins on a desktop and continues elsewhere. Then the May 21, 2026 release notes added appshots and goal mode, which pushed Codex further away from generic prompting and closer to task handoff. Together, those updates make Codex feel less like a chat interface and more like a working surface for live context.

The practical shift is simple: instead of describing a screen in words, you can attach the screen itself. Instead of asking for “help with this,” you can define a result, constraints, and the kind of output you want. That reduces the friction of starting, and it also reduces the chance that you spend the first five turns rebuilding context that was already visible on your device.

What appshots are and when to use them

Appshots let you attach an app window to a Codex thread with a hotkey, so Codex can work from the actual content you are looking at. That is useful when the task depends on layout, visible labels, a specific draft, or a set of details that would be tedious to transcribe. The point is not to hand Codex your whole desktop; the point is to capture the minimum screen context needed for the task.

Use appshots for slides, docs, browser tabs, dashboards, code editors, and interview prep notes. Those are all cases where the screen already contains the raw material Codex needs to summarize, rewrite, compare, or organize. If you only need one window, attach one window. If you only need a single document section, avoid cluttering the thread with unrelated apps or personal workspace.

A good rule is to ask yourself what Codex must see in order to finish the task correctly, then include only that. A focused appshot makes the task easier to scope, easier to verify, and easier to repeat later. It also gives you a cleaner review process because the output can be judged against a narrower, more relevant source of truth.

Set a goal Codex can actually finish

Goal mode works best when you specify the destination before you start. Write down the desired outcome, what success looks like, and any constraints that should not be ignored. A vague request like “clean this up” leaves too much room for interpretation; a goal like “turn this meeting note into a 150-word recap with action items and no changes to the names or dates” gives Codex something concrete to complete.

Goal mode is especially useful for multi-step work such as summarizing notes, drafting a response, or debugging a small workflow. In those cases, the model needs to move through several decisions in sequence, and a clear finish line helps prevent drift. Include tone, length, format, and what not to change, so the output stays useful without forcing a long correction loop.

Think in terms of a brief task spec. What should the output be? Who is it for? What must stay untouched? When those pieces are explicit, Codex can do more of the drafting and structuring work on its own, and you can spend your time reviewing the result rather than redefining the assignment.

Build a leave-and-return workflow on desktop and mobile

The new workflow is designed for moments when you do not want to babysit the task. OpenAI’s May 14, 2026 update made it possible to stay connected to active Codex work through the ChatGPT mobile app, including work that begins on a Mac. That makes it easier to hand something off, step away, and come back later without losing the thread.

When eligible, locked computer use can help longer tasks continue after the Mac locks. That changes the rhythm of work in a useful way: you can start a task with a clear goal, review approvals, diffs, screenshots, and test results from mobile, then return to the desktop only when the task needs a final human decision. It is a better fit for real work than a workflow that demands constant supervision.

The safest way to use that setup is to reserve the handoff for bounded tasks you can review. If Codex is changing files, generating comparisons, or running checks, keep the review step explicit. Mobile is the place to monitor progress and approve or reject what you w

Sources