AI to manage AI? Claude creator says prompt engineering is ending too, just like software engineering. Reveals what's next

The AI world may already be moving on from prompt engineering.

Just months after claiming that "software engineering is dead," Anthropic co-founder and Claude Code creator Boris Cherny has a new prediction: humans won't be writing prompts for AI much longer.

Instead, he says the future belongs to "loop engineering" — a new approach where AI agents write, refine, and manage prompts on their own while humans simply define the goal.


"It's an agent that prompts Claude. I don't write the prompt anymore. Claude writes the prompt," Cherny recently told Business Insider.

In other words, AI is starting to manage AI.

So what exactly is a loop?

Think of a loop as a self-running AI workflow.

Instead of feeding an AI model one instruction after another, users can give it a broad objective and let it keep working until the task is complete.

A command such as "/goal" can tell an AI system to continue solving a problem, checking its work, creating new prompts, and improving results without needing constant human guidance.

The idea is catching on fast among AI insiders.

Peter Steinberger, creator of OpenClaw, says developers should stop obsessing over prompts and start building systems that generate prompts automatically.

"You shouldn't be prompting coding agents anymore. You should be designing loops that prompt your agents," he said.

Google Cloud executive Addy Osmani agrees. According to him, effective loops combine automation, plugins, connectors, worktrees, skills, and specialized sub-agents working together.

AI agents are becoming co-workers

The biggest shift may be how people think about AI.

Instead of using AI like a tool, experts increasingly suggest treating it like an employee.

Claire Vo, founder of ChatPRD, says users should imagine they are onboarding a new hire.

"This is the time for the manager. You are designing a job," she explained.

In these systems, one AI agent might write code, another could review it for bugs, while a third manages documentation or deployment.

The human's role? Supervision.

There's a catch: AI workers aren't cheap

While loop engineering promises less work for humans, it can mean bigger AI bills.

Every agent consumes tokens, and running multiple agents simultaneously can become expensive quickly.

Steinberger recommends scheduling agents to work hourly or daily instead of constantly running in the background. Osmani also warns that companies should only deploy additional sub-agents when the benefits justify the cost.

In short, the AI workforce may need a budget just like a human one.

Boris Cherny is also tired of the term 'vibe coding'

Cherny has another complaint: the phrase "vibe coding."

The term, coined by OpenAI co-founder Andrej Karpathy in 2025, became one of the hottest buzzwords in tech and was widely used to describe AI-assisted programming.

But Cherny says the label no longer fits.

Today's AI coding tools are generating millions of lines of production software and helping create billion-dollar businesses. Calling that "vibe coding," he argues, seriously understates what's happening.

He even asked Claude to suggest alternatives.

One option was "agentic engineering" — another term linked to Karpathy — but Cherny isn't sold on it.

For now, he's crowdsourcing ideas and inviting people to suggest better names.

One thing, however, seems increasingly clear: prompt engineering may already be yesterday's skill.

The next big thing in AI could be learning how to manage teams of autonomous digital workers that manage themselves.