Action Mode + Workshop Mode

Level 1 · Course 7

Here's a situation you know. You're talking through a problem with your agent — trying to flesh it out, bounce ideas around, figure out what's actually broken. Mid-sentence, your agent disappears. Thirty seconds later it comes back with a "fix." It wasn't listening. It doesn't understand the full scope of the problem. It put a Band-Aid on a bullet wound. Now you have to undo whatever it built and start the conversation over. You didn't ask it to build anything. You were just talking.

Workshop mode kills all execution. Your agent can read internal files for reference, but it cannot write, cannot deploy, cannot search the web, cannot build anything. It sits in its chair and listens. You bounce ideas. You think through the full action plan together. No premature fixes. No Band-Aids. This also stops drift. Hard rules that keep your agent from executing prematurely — rules that hold at 50K tokens — get ignored at heavier levels of context. Workshop mode stops drift in its tracks because the agent cannot execute at all. It's not a rule the agent can fade on. It's a structural lock.

Workshop mode is read-only by default, but you can curate your own version — add allowances for the tools your agent can use while workshopping. Some operators allow web search. Some allow file reads only. It's your gate, tuned to how you work. Then you say "action mode" and the release comes. All tools back on. Back to your regular scheduled productivity. Action mode is the default operating state of your agent. Workshop mode is a sidebar conversation — a space to think before you build. This is the only hard stop in the system. Nothing happens until you say the words.