Boris Cherny created Claude Code. Every quote below is from Anthropic's video "Reflecting on a year of Claude Code" (June 2026), with the minute it's said so you can check it. The commands are from the Claude Code docs as of October 2026.
Skip plan mode
"I just do auto mode for everything, because then I start my Claude, it starts to work, and then I just move on to the next Claude." (7:13)
Press Shift+Tab until the status bar under the prompt says "auto mode on". In auto mode a second model checks each action for safety, so you aren't approving every command yourself.
On Claude Code v2.1.283 or later, auto mode is already the starting mode in the terminal and VS Code. If yours still starts in plan or manual mode, update Claude Code or run /doctor, which offers to switch your default.
Get the rest of this guide free
Enter your email once and every guide and setup on the site opens, this one included. You'll also get AI This Week by MattyMoe: the week's AI news for business owners, every Friday.
Write the fix down once
"Every single time Claude makes a mistake, I don't tell Claude to do it differently. I tell it to write it to the CLAUDE.md, or to make a skill." (0:54)
The next time you correct Claude, add this line so the fix sticks for every future session.
Write this to CLAUDE.md so you don't make that mistake again.
Type /memory to open your CLAUDE.md files and read or edit them yourself.
Make it check its own work
"When we talk about verification for agents, it's something slightly different. It's like, can the agent run the thing?" (1:40)
Give Claude a way to run what it built: the test command, a local dev server it can open, or a browser it can click through. Then set a finish line with /goal and Claude keeps working until the condition is met.
/goal the signup page loads and the form submits without errors
Talk to a loop
"I don't talk to an agent anymore. I talk to a loop or I talk to a routine and it prompts Claude for me." (10:56)
/loop runs a prompt again and again while the session stays open. /schedule sets up a routine that runs in the cloud on a schedule, even when your laptop is closed.
/loop 30m check for new failing tests and fix them
A real routine from the Claude Code team (5:31) finds bug reports nobody has answered in 5 hours and puts up a fix.
Give it less
"You give it the minimal possible system prompt, the minimal possible tools, and then you let the model figure it out." (16:29)
Type /context to see what is filling Claude's context window as a grid. Type /doctor (also /checkup) to check your setup for extension and CLAUDE.md problems. It asks before it changes anything.
Run several at once
"Probably half my engineering now I do on my phone." (15:00)
Boris went from six terminal tabs to one agent view (14:30). Type /remote-control (or /rc) to make a session available from claude.ai and the Claude phone app, then start your next task in a new session instead of waiting on the first one.