ChatGPT now has Chat, Work and Codex. In this video Lukas gives a practical way to decide which mode makes sense and when Work only adds unnecessary complexity.
The value of the video is the sorting. Many people see ChatGPT, Work and Codex next to each other and ask whether this is all the same thing with different labels.
The answer in the video is deliberately simple: Chat is for quick answers and thinking together. Work is for tasks where you want a finished output. Codex is for software development, repositories, terminals, tests and debugging.
The important question is not whether Work looks impressive. The useful question is whether it actually takes work off your plate or whether you are starting a large process for a small question.
| Mode | Use it for | Not for |
|---|---|---|
| Chat | Quick answers, ideas, rewrites, back-and-forth | Long tasks with multiple files and a finished deliverable |
| Work | Delegated tasks, research, documents, spreadsheets, presentations, reports | Every small question just because the mode is new |
| Codex | Codebases, terminals, tests, debugging, real repository changes | General research or pure text work without a codebase |
| Claude Co-work | A useful comparison point for delegated work and final outputs | A permanent truth about which tool is always better |
ChatGPT Work is for longer tasks where you want a finished output, such as research, documents, spreadsheets, presentations, reports or small Sites.
No. Work is for general projects and deliverables. Codex is still the right place for codebases, terminals, tests, debugging and repository work.
For quick answers, rewrites, brainstorming, image ideas and fast back-and-forth, normal Chat is usually faster and more efficient.
Longer agentic tasks can count differently from short chat messages. If Work and Codex draw from the same agentic usage structure, Work should be used deliberately.
This page currently uses an editorial chapter summary. The subtitle files are prepared separately for YouTube.
The opening sorts out the new ChatGPT feeling: there is no longer just one chat, but different working modes. Lukas sets the simple model: ask Chat, delegate to Work, use Codex for code.
This is the core distinction. Chat is fast and conversational, while Work takes more context and tries to deliver a finished output.
Work is framed as an interface for tasks with files, research and multiple steps. Not magic, just delegation with more persistence.
The section separates Work from normal Chat. If you use Work, you should expect an outcome, not just the next reply in a conversation.
Codex remains the better choice for repositories, terminals, tests, debugging and real code changes. Work can do a lot, but code is not its cleanest center.
Lukas connects the topic to the website buildout. Individual videos become pages and topic areas, so useful content does not only live on YouTube.
This is the practical cost point: long agentic tasks are not the same as quick chat questions. Anyone using Work heavily has to treat limits and effort seriously.
For fast questions, small rewrites or spontaneous thinking, Work is often overkill. Normal Chat is still the better move for many cases.
Work makes sense when several steps, files or sources need to become one result. That is where delegation becomes the actual benefit.
The assessment stays sober. Work can help, but quality comes down to how often you need to intervene and how useful the finished output is.
The bigger point: AI use is moving from chat answers to delegated work. That makes it more important to choose the right mode.
Lukas compares Work with Claude Co-work by working feel and output, not as a fanboy comparison. The verdict stays open because these tools change quickly.
The closing gives the simple decision rule: Chat for answers, Work for outcomes, Codex for code. That clarity is the value of the video.
I work 1:1 with freelancers, consultants, coaches and small teams on practical AI workflows, automation and custom tools.
Book a free strategy call