The 60-second version
An AI assistant is a tool you talk to in plain language, like a fast, very literal colleague who has read a great deal but knows nothing about your job until you tell it. It can draft, summarise and suggest, but it does not decide, and it can be confidently wrong.
A prompt is simply what you type or say to the assistant: your request, plus any background and examples you give it. The clearer and more specific the prompt, the more useful the answer, so treat it like briefing a new starter, not typing a search.
A context window is how much the assistant can hold in mind at once: your instructions, the document you pasted, and the conversation so far. When a chat gets long it starts to forget the earliest parts, so for a big task give it the key facts again rather than assume it remembers.
Start here
The useful move is not to adopt everything. Pick one repeated workflow, define the human approval point, and test whether the new surface reduces accepted work, not just demo friction.
See the detail and the sources →A stronger model is still only useful if it clears your task. Keep the same input, standard and reviewer before you switch a workflow.
See the detail and the sources →Routing is now part of the work. Decide which tasks need the stronger route, which need speed, and which still need a person before anything changes outside the workspace.
See the detail and the sources →
The one that mattersOpenAI's launch day was not just a list of features. It described a different way work will happen: agents wait between conversations, plugins bring app surfaces into the conversation, Codex continues in cloud workspaces, and shared Pages or Spaces carry the work forward. That can save time, but it also moves AI closer to real operations, where a wrong draft, route, permission or action matters.
What's genuinely new
OpenAI says it announced more than 20 changes across ChatGPT, Codex, models, agents, plugins, collaboration and the API, including Dots, GPT-6.1 Sol, Codex Cloud, Decisions API, Agents API computer use, MCP Events, Pages, Meetings and Pro 500.
What the makers claim
The provider frame is capability and momentum: more surfaces, more autonomy, more integration. That is not the same as proof that any one team's workflow becomes safer or cheaper on day one.
Where it helps
A repeated job where the input is approved, the possible output is clear, and the person can review the draft, route or decision before it leaves the workspace.
Where it doesn't
A vague responsibility, sensitive data flow, irreversible action, live client communication or production workflow where nobody has written the approval rule yet.
OpenAI's DevDay 2026 recap says it announced more than 20 major changes across ChatGPT, Codex, models and new ways of working with AI.
Choose one workflow before choosing a tool: meeting to action list, project event to plan, source pack to Page, code diff to risk list, or customer request to finite routing decision.
Try this
A repeated internal task keeps coming back: meeting notes, project updates, customer triage, code review, source gathering or weekly reporting.
Prepare the first pass, gather context, suggest a route, update a shared draft or list missing information without sending, publishing or changing anything final.
Choose the workflow, approve the sources, decide the acceptance standard, check the output and approve any action that leaves the workspace.
Pick one repeated task and write the output you would accept
List the source material AI may use and the material it must not see
Write the approval step before choosing Dots, Agents API, plugins, Pages, Space or Codex
Run the workflow on copied or non-sensitive material
Score accepted output, correction time, source errors, privacy issues and saved time before rollout
Nothing gets sent, published, paid, deleted, routed to a real customer or changed in production until a named person approves it.
The Human Lens · what stays yours
The more an AI surface can watch, draft, route and act, the more clearly the person has to define what good means. DevDay's safest lesson is not that AI can do more. It is that the review step has to move closer to the workflow before the workflow moves closer to the business.
Worth watching
One question to carry into your week
See you next Monday.