What Did Boris Cherny Actually Say?
During a fireside chat at Fortune's Brainstorm Tech conference in June 2026, Cherny described how his relationship with AI has evolved through three stages:
- Writing code by hand — the traditional way
- Prompting Claude to write code — the first AI shift
- Writing loops that prompt Claude automatically — where he is now
He also revealed he hasn't written a single line of code by hand in eight months. Claude Code — Anthropic's flagship coding agent — is itself 100% written by Claude Code. The system is writing itself.
The exact quote: "My coding was prompting Claude to write code. Now it's leveled up again to the next abstraction where I don't prompt Claude anymore. I have loops that are running. They're the ones prompting Claude and figuring out what to do. My job is to write the loops."
What Is a Loop?
A loop is an automated system that runs repeatedly — it prompts an AI, evaluates the output, decides what to do next, and prompts again. The human doesn't participate in each cycle. They design the system and let it run.
Instead of a human saying "write this email" and then reviewing and editing it, a loop says "monitor our inbox, identify messages that need a response, draft replies in our brand voice, flag anything unusual for human review" — and runs continuously without the human being in each cycle.
Why This Matters for Prompts
Here's the thing that the "prompts are dead" headline gets wrong: loops are full of prompts. They're just prompts that run automatically, not prompts that a human types each time.
The skill of prompt engineering doesn't disappear in a loop-engineered world — it moves up a level. Instead of writing a prompt for a task, you write prompts for a system. The prompts need to be more precise, more robust, and more self-contained because no human is there to clarify or correct.
What this means for the Promptaholics library: Prompts aren't becoming less valuable — they're becoming building blocks for automation. A well-structured CoSTAR prompt doesn't just work when you type it. It works when a loop runs it 500 times a day. That's the next frontier we're building toward.
The Five Parts of a Loop
Engineer Addy Osmani's influential post on loop engineering identified five components that most production loops share:
A Real-World Loop Example
Here's a loop that a content marketer might build — no code required, using tools like Zapier or Make:
How to Start Thinking in Loops
You don't need to be a developer to apply loop engineering principles. The mindset shift is simpler than the technical implementation:
- Identify repetitive AI tasks. Any task you do with AI more than once a week is a candidate for a loop.
- Define the trigger. What event starts the process? A new email, a new file, a scheduled time, a data change?
- Write the prompts first. Before building any automation, write the prompts that would do each step manually. Get them working perfectly by hand before you automate them.
- Add verification steps. What does "good output" look like? Define it explicitly so the loop can check its own work.
- Start with one loop, one task. Don't try to automate everything at once. Pick the highest-value repetitive task and build one loop for it.
The Skill That Doesn't Change
The thing Cherny's statement actually reveals is that the skill of precise, structured prompting is more important than ever — not less. When a human reviews every AI output, a mediocre prompt is just an annoyance. When a loop runs that prompt 10,000 times without human review, a mediocre prompt is a disaster.
Loop engineering doesn't make prompt engineering obsolete. It makes prompt engineering critical infrastructure.