FAQ

No. Your session history lives in .jsonl files on your local disk, not in your Claude or OpenAI account. Switching accounts, or switching between Claude Code and Codex, leaves those files alone. Open the project folder later and the tool finds your prior sessions where they were.

Conversation history is still the wrong thing to rely on across weeks. Have the agent write the important stuff into markdown files as it works: what's done, what's next, open questions, decisions and why. Next session any model you open the folder with can read the markdown and keep going, even with no memory of the previous one.

For the browser tools (ChatGPT, Claude, Gemini), no. You type in plain language and get answers back.

For the terminal tools (Claude Code, Codex CLI), you'll pick up what you need as you go: how to run basic commands, paste errors back, and read a diff before approving it. You don't need to learn to code: the agent writes it. The skill is describing your problem clearly and asking good follow-up questions. That is a research habit, not a programming one.

Assume it will, and set up your workflow so catching it is cheap. The hallucinations that hurt research work are the confident-looking ones: a paper that doesn't exist, an API that was never released, a function signature that looks right but isn't.

Give the agent a way to ground its claims in real data: an MCP server for the database you care about, a file it can read, a test it can run. Ask for citations or file:line references for anything factual; if it can't point to a source, treat the claim as a guess. Never take a generated number at face value. If it reports "23 % improvement," ask it to show you the row.

Anthropic's own Claude Code guide puts it as "give Claude a way to verify its work." That is the whole rule.