Was this newsletter forwarded to you? Sign up to get it in your inbox.
We talk about AI progress like it’s a race between models. That’s the part with version numbers, and what you argue about over dinner. But the model is only one piece. It reaches your work through the app you open it in, the files and tools you let it touch, the instructions you’ve saved from last time, and the person keeping all of that from going stale. Any of those can change your results as much as a new model would—sometimes more.
We rewrote two of our guides this week. The models didn’t change. The infrastructure around them that governs their behavior and capabilities did. These guides are written as much for agents as for humans, so share them with your agent and ask it: “What can we learn from this that we can adapt for how we work?”
Knowledge work has a home in ChatGPT now
When we initially wrote our Codex for Knowledge Work guide, the idea of using a coding-agent interface for something other than coding was so new that even OpenAI hadn’t built a dedicated home for it. Now it has. Quick questions stay in Chat, longer assignments move to Work, and software jobs go to Codex. For knowledge workers who don’t touch code, the vast majority of your tasks can be handled by Work.
We’ve retitled the guide ChatGPT for Knowledge Work and rewritten it around that new setup. The guide keeps its 16 workflows, but recasts most of them for Work. The update also covers several features that didn’t exist when we wrote the original guide:
- Goals: Type /goal to turn an objective into a persistent goal with a definition of done.
- Projects: Choose between ChatGPT projects, which keep cloud conversations, files, and instructions together, and local projects, which give Work or Codex access to a folder on your computer.
- Scheduled work: Use Scheduled Tasks for recurring jobs in Work and thread automations for longer loops in Codex.
- Browser access: The app’s built-in browser has its own signed-in profile. Choose Chrome when a task needs your existing sessions, tabs, or extensions.
Most AI tools stop at the summary
Research is the easy part. The work is everything after: the report drafted, the spreadsheet updated, the recap posted in Slack, the follow-up actually sent. Grok Bot’s Bots do that part. Each one runs on its own cloud computer, signed in to your CRM, your ATS, your inbox, clicking and typing through the same software your team already uses. Give a Bot a job instead of a prompt, then check the result the way you’d check a coworker’s. Built by the team behind SpaceXAI.
Our OpenClaw recommendation changed
Our first OpenClaw guide focused on the appeal of a personal agent. A Claw lives in your messaging app, connects to tools you authorize, and keeps doing recurring jobs without a fresh prompt. For someone willing to maintain it, that setup can still be unusually useful.
The update adds what we learned from running Claws ourselves. A stronger model still can’t log in when a credential expires. It can’t notice that an integration broke last week, or that a job you set up months ago should’ve been turned off. Putting the agent on a server keeps it running—but somebody still has to keep its tools connected, its permissions current, and its memory from filling up wtih stale instructions. For now, that somebody is a burden.
That maintenance load only grows as a company adds people, so we’re trying something different with the Every Agent. It lives in Slack, so the whole company shares one agent—but each person works through their own connections and own context.
A personal Claw can still make sense if you have recurring work specific to you, want control over its tools and instructions, don’t need company context, and are willing to maintain it. The updated guide helps you decide whether that tradeoff is worth it, then explains where Claws tend to break and how to limit their authority.
Katie Parrott is a staff writer at Every. You can read more of her work in her newsletter. To read more essays like this, subscribe to Every, and follow us on X at @every and on LinkedIn.
Everyone’s a builder now. Every All Access gets you the full membership plus the Builder Pack—$9,000+ in credits for the tools we build with.
