Boris Cherny - the person who literally built Claude Code - just posted his 10 tips for using it. I read through them nodding the entire time.
I’m a comedian. I don’t write code.
Well - I didn’t used to. Now I kind of do? But the point is: I started using Claude Code for non-coding work, and I apparently arrived at the same workflows as the guy who made the thing.
Which either means I’m a genius or these patterns are just…obvious once you start using it. Probably the second one. But Dario if you’re reading this plz lmk.
ok here’s the thing though - Boris’s tips aren’t really about coding. They’re about how to work with an AI agent. And that’s a communication skill. Guess which one comedians are better at.
Let me walk through all 10.
1. parallel sessions
Boris says spin up 3-5 git worktrees at once, each running its own Claude. It’s his team’s #1 productivity tip.
I do this too. Except mine look like: one session is drafting a newsletter, one is analyzing my email marketing data, one is being my therapist about a stressful client text, and one is building me an invoicing tool.
Same principle. Wildly different vibes.
2. plan mode first
Boris says pour your energy into the plan so Claude can one-shot the implementation.
I do this for everything - not just code. “Here’s what I need to accomplish today, here’s the context, here are the constraints.” My Claude sessions start with a briefing, not a task. The prompt IS the work. Rush the prompt, get garbage. Spend 10 minutes writing a clear brief, Claude nails it first try.
Someone on Boris’s team has one Claude write the plan, then spins up a SECOND Claude to review it as a staff engineer. I do the same thing but with writing - one Claude drafts, another one tears it apart.
3. invest in your CLAUDE.md
This is the big one. Boris says after every correction, tell Claude: “Update your CLAUDE.md so you don’t make that mistake again.”
My CLAUDE.md is - and I say this with full awareness of how unhinged it sounds - one of the most important documents in my life right now.
It includes:
- My banned writing phrases (never say “delve,” never say “seamless,” if you write “in today’s world” I will close the laptop)
- How to handle when my producer texts me late at night. Literally: “Remind Alex the reply is probably simpler than his body thinks it is”
- My distribution channels, because I keep forgetting I have warm contacts who’d share my stuff
- File organization rules
- Idea capture workflows
- Daily work logging instructions
I stopped thinking of it as a config file months ago. It’s closer to a working relationship. You’re teaching the AI how to be useful to YOU - not to some generic user.
The non-coding version might actually be more powerful than the coding version? Because the stuff it captures is so personal. Your stress patterns. Your bad habits. Your voice.
4. custom skills
Boris says create reusable skills and commit them to git. If you do something more than once a day, make it a skill.
I have 19 custom slash commands:
- /chief-of-staff - daily planning, syncs meeting transcripts, runs a morning sweep
- /life-cfo - financial advisor that knows my contracts and runs real numbers
- /humanizer - catches AI-sounding writing before I publish
- /branding-coach - channels a specific branding framework and pushes back on weak positioning
- /publish-everywhere - content distribution across platforms
- /morning-sweep - automated daily organization
I’m not a developer. I built all of these with Claude Code. I described what I wanted and iterated until it worked.
A “skill” is really just a process you’ve done enough times to name. Once you name it and write it down, Claude can run it for you. That’s it.
5. bug fixing
Boris says paste a Slack bug thread into Claude and just say “fix.”
I paste a confusing client email and say “what is this person actually asking me and draft a response.”
Same thing. The bug is social anxiety.
6. better prompting
Boris’s team does this thing where they challenge Claude - “Grill me on these changes.” “Prove to me this works.”
I use: “Read this draft and tell me where I’m being lazy.” And: “What would someone who disagrees with this say?”
Works for writing, business decisions, pricing strategy, anything where you need someone to push back on you. Most people prompt AI like it’s an assistant. The better move is to prompt it like a sparring partner.
7. environment setup
Boris’s team loves Ghostty terminal, uses statuslines, color-codes their tabs, and - this one’s big - uses voice dictation because you speak 3x faster than you type.
I have all of this. Plus an “agent bus” - a file-based system where my Claude sessions leave notes for each other. One session finishes work, logs what it did, and leaves context for the next one. My chief of staff session checks the bus every morning.
When I describe this to engineer friends they say “oh so you built a message queue.” Sure. I built a message queue. I’m a comedian who built a message queue.
8. subagents
Boris says append “use subagents” when you want Claude to throw more compute at a problem.
I use these for research. One agent digs into a topic while my main session keeps working. It’s like having an intern who’s also a genius but has no long-term memory.
9. data & analytics
Boris’s team uses Claude to query BigQuery. He says he hasn’t written SQL in 6 months.
I’ve had Claude pull and analyze my email campaign data through APIs. Open rates, click patterns, which subject lines worked. I don’t know SQL either, but I don’t need to - I just tell Claude what I want to know and it figures out how to ask the database.
10. learning mode
Boris says enable “Explanatory” output style and have Claude explain the why behind its changes.
My most-used prompt pattern is basically this: “Explain this to me like I’m smart but don’t know this specific thing.” I’ve learned more about APIs, email marketing infrastructure, and terminal workflows in the last year than in the previous decade. Not from courses. From building stuff with Claude and asking it to teach me as we go.
so what does this actually mean
I keep coming back to this: the skills that make someone good at using AI for coding are the same skills that make someone good at using AI for anything. Clear communication. Good documentation. Breaking work into parallel streams. Building reusable processes. Investing upfront in context and planning.
These are thinking skills. They just happen to live inside a terminal.
Non-technical people might actually have an edge here? We’re not tempted to over-engineer things. We describe what we want in plain language. We think in workflows and outcomes, not implementations.
Boris’s 10 tips are accidentally a universal guide to working with AI agents. The fact that they came from an engineering context is almost a distraction.
If you were going to write a CLAUDE.md for your life - your daily work, your creative process, your bad habits, your stress patterns - what would be in it?
I’m genuinely collecting answers. Reply and tell me.
I teach a course about this called Code for Creatives - non-coders using Claude Code for exactly this kind of stuff. If this resonated, that might be your thing.