Two developments from the past 48 hours that you'll notice if you work with AI daily. OpenAI is merging the usage limits of ChatGPT and Codex, and DeepSeek is releasing an experimental coding agent that reads along in your development team's chats.
ChatGPT and Codex: one usage budget from December
From December, ChatGPT and Codex will share one usage budget, measured in tokens, on the Plus, Pro and Business subscriptions. Bright reports this based on OpenAI's help centre. Larger models and complex tasks will use a bigger share of that budget. Conversations in Instant mode don't count. Features from ChatGPT Work, intended for larger tasks, are moving to the regular ChatGPT. OpenAI has not yet disclosed how large the shared budget will be.
At the same time, OpenAI is limiting the number of paid accounts per device and the number of subscriptions per payment card. This applies to Go, Plus and Pro, not to Free, Business, Enterprise and Edu. According to OpenAI, anyone who needs more room should buy credits or use the available resets rather than create extra accounts. OpenAI does not state the maximum numbers.
What does this mean for you? If you use ChatGPT for texts and Codex for code on the same subscription, a long programming session can fill up your chat allowance. Keep routine questions in Instant mode and monitor your usage once the change takes effect. The current subscriptions and prices are on the ChatGPT page; for what you may and may not put into ChatGPT in your field, see for example the GDPR check for ChatGPT in software development. If you're unsure whether a different subscription would suit you better, compare models.
DeepSeek Super-Agent: a coding agent that doesn't wait for you to ask
The Harness team at DeepSeek has published Super-Agent, an open-source prototype of an autonomous programming agent. According to Techzine, the agent can follow developers' group chats, recognise programming problems there and suggest solutions, even when nobody asks it to. The Next Web reports that the code is available on GitHub under the MIT licence as dsh-superagent-prototype. The agent works on a single GitHub repository, listens in on Feishu, ByteDance's work chat, and replies in the original thread, for example with a pull request. DeepSeek explicitly calls it a research prototype and not an official product; the public code differs from the internal version.
What does this mean for you? For most Dutch teams this is not yet a usable product: it is built around Feishu, and you need to review and adapt the code yourself before you deploy it. The idea is worth watching, though: an agent that reads along in your team chat also sees everything that passes through it, including client names and internal agreements. Only deploy such an agent with limited permissions, and agree what is allowed to appear in that chat. For how DeepSeek scores on privacy, see the GDPR check for DeepSeek in software development. Established coding assistants are listed under AI for software development, such as GitHub Copilot and Claude. For the difference between the models, see Cursor vs DeepSeek.
In brief
- From December, ChatGPT and Codex share one token budget on Plus, Pro and Business; Instant conversations do not count towards it.
- OpenAI limits the number of paid accounts per device and subscriptions per payment card.
- DeepSeek Super-Agent is an open-source research prototype (MIT) that reads along in team chats itself and responds to pull requests.
- All models and their privacy scores are at models and safety.
