What is the difference between the inner loop and outer loop that Ronacher describes?

The inner loop is where an AI agent executes work—using tools, reading files, testing code, and learning from results before stopping. The outer loop (harness) is the human-controlled framework where engineers define the task, set safeguards, and monitor whether the AI has achieved the goal.

How are engineers currently implementing this loop-based approach in practice?

Early adopters are organizing tasks in Markdown files (PLAN.md, TODO.md) that AI agents can read and execute, and using project management tools like Linear to track tasks designed for loop-based workflows. Some teams are also experimenting with checklist-style tracking to communicate loop boundaries to AI systems.

Back to articles

Software engineering shifts to "loop" design as AI coding agents reshape development

ITmedia AI＋2d ago4 min read

Key takeaway

Software engineering is undergoing a structural shift as AI coding agents become more capable. Rather than writing code themselves, engineers are increasingly designing "loops"—outer frameworks where humans set tasks and review results, with inner loops where AI agents execute the work. This transformation parallels broader changes in how people collaborate with AI, though its practical limits and security implications are still being tested by early adopters.

Summaries like this, in your inbox every morning.

3 Key Points

What happened
Armin Ronacher, creator of the Flask web framework, has written about "The Coming Loop," distinguishing between an inner loop (where AI agents execute tasks) and an outer loop or harness (where humans oversee the process). This framework reflects a fundamental shift in how software engineers structure work with AI coding assistants.
Why it matters
As AI coding agents become more capable, engineers are rethinking their role. Rather than writing code directly, they increasingly manage loops—defining tasks, reviewing outputs, and deciding when the AI has completed its work. This mirrors how the relationship between humans and machines is evolving across industries, potentially making loop engineering a core skill rather than traditional code writing.
What to watch
The article highlights concrete practices emerging now: engineers are using Markdown files (PLAN.md, TODO.md) to communicate task lists to AI agents, and tools like Linear are being adapted to track loop-based workflows. Early adopters are experimenting with whether this approach can scale, though questions remain about security, cost, and whether AI systems can be trusted to operate reliably within defined boundaries.

FAQ

What is the difference between the inner loop and outer loop that Ronacher describes?: The inner loop is where an AI agent executes work—using tools, reading files, testing code, and learning from results before stopping. The outer loop (harness) is the human-controlled framework where engineers define the task, set safeguards, and monitor whether the AI has achieved the goal.
How are engineers currently implementing this loop-based approach in practice?: Early adopters are organizing tasks in Markdown files (PLAN.md, TODO.md) that AI agents can read and execute, and using project management tools like Linear to track tasks designed for loop-based workflows. Some teams are also experimenting with checklist-style tracking to communicate loop boundaries to AI systems.

Discussion

No discussion yet for this article

Stay ahead with AI news

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.

Get Started Free

Free · takes 30 seconds · unsubscribe anytime

1 minute a day. The AI essentials.

200+ sources · Email / LINE / Slack

Get it free →