TL;DR
Anthropic’s Claude Code team published a plain definition of agentic loops on June 30, 2026: repeated agent work cycles that run until a stop condition is met. Thorsten Meyer AI’s July 1 dispatch frames the four loop types as a delegation ladder, showing what users stop doing at each level of autonomy.
Anthropic’s Claude Code team has published new guidance on agentic loops, and Thorsten Meyer AI is framing the model as a four-rung delegation ladder for deciding how much work to hand off to AI systems. The framework matters because it gives developers and business teams a practical way to separate a tool they operate from a process that can run with less direct supervision.
The dispatch says Anthropic defines a loop as an agent repeating cycles of work until a stop condition is met. The Thorsten Meyer AI framing adds a business-facing lens: each loop type is defined by what the human stops doing, from checking work manually to no longer writing the starting prompt.
The four rungs described are turn-based skills, goal-based loops, time-based loops, and proactive workflows. In the first, a user still starts the task, but a skill can encode verification steps so the agent checks its own output. In the second, a goal command lets a separate evaluator model keep the agent working until the stated condition is met or a turn cap is reached.
The higher rungs shift more responsibility away from the user. With time-based loops, a schedule or interval starts the work. With proactive workflows, events can trigger tasks without a human prompt in real time, according to the dispatch. The source material says some features are research previews, and it attributes the definitions, primitives, and examples to Anthropic while identifying the delegation ladder framing as the author’s interpretation.
The delegation ladder: four agentic loops, and what each lets you stop doing
Strip the hype and a “loop” is simple — an agent repeating work until a stop condition is met. The useful lens isn’t the mechanics, it’s what you hand off. Four loop types = four rungs of delegation, from a tool you operate to a process that runs.
The whole framework reduces to one question about your own work: where am I the bottleneck, and which single piece can I hand off? Can you write the check? Is the goal concrete? Does the work arrive on a schedule? That answer picks your rung — and you climb one step at a time. The real skill isn’t operating a loop; it’s the judgment of what to delegate and how far — enough hands off to gain leverage, enough on the wheel that “runs without you” doesn’t become “runs away from you.”
Autonomy Becomes A Design Choice
The main effect of the framework is that it turns AI autonomy into a decision that can be scoped. Instead of asking whether agents should be trusted broadly, teams can ask a narrower question: should the system take over checking, stopping, starting, or prompting?
That distinction matters for cost, quality, and control. The dispatch warns that autonomy is metered, so longer or recurring loops can raise usage unless teams use clear goals, capable lower-cost models, turn limits, and scripts where reasoning is not needed. It also stresses that output quality depends less on the loop itself than on the surrounding system: clean codebases, reusable skills, fresh-context review agents, and fixes to the process rather than one-off corrections.
AI automation workflow tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Anthropic’s Loop Guidance Explained
The source material ties the dispatch to Anthropic’s Getting started with loops, credited to Delba de Oliveira and Michael Segner on the Claude blog on June 30, 2026. The dispatch appeared in Insights AI Dispatch on July 1, 2026.
The lower rungs are aimed at shorter or clearly bounded work. A turn-based skill can verify a UI change by starting a dev server, clicking controls, checking screenshots, reviewing console output, and running a performance trace. A goal-based loop works best when the success condition is measurable, such as passing tests or reaching a performance score, because the evaluator has less room to accept a vague result.
“a loop is an agent repeating cycles of work until a stop condition is met”
— Anthropic’s Claude Code team, as summarized by Thorsten Meyer AI
task automation software for business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Availability And Adoption Remain Open
Several details are not settled in the source material. It does not say how widely each Claude Code primitive is available, how many users are testing the research-preview features, or what adoption looks like outside developer workflows.
It is also unclear how teams will set safe stop criteria for work that is less measurable than tests, scores, or scheduled maintenance. The dispatch presents the ladder as a way to choose delegation levels, but it does not provide independent evidence on failure rates, cost ranges, or reliability across large-scale agent runs.
AI process management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Teams Test The Right Rung
The next step for readers is practical evaluation. The dispatch argues that teams should start with the simplest working loop, identify where humans are the bottleneck, and climb only one rung at a time when the task supports it.
For Anthropic, the next milestone is likely broader feedback on how developers use skills, /goal, /loop, /schedule, and proactive workflows in real projects. For businesses, the near-term test is whether these patterns reduce manual oversight without creating runaway costs or weak quality checks.
AI task scheduling software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the main news here?
Anthropic published guidance on agentic loops, and Thorsten Meyer AI framed the four loop types as a delegation ladder for deciding what AI systems can take over.
What are the four agentic loops?
The four described rungs are turn-based skills, goal-based loops, time-based loops, and proactive workflows. Each shifts one more part of the work from the human to the agent.
Why does this matter for businesses?
It gives teams a clearer way to decide how far to delegate. The framework links autonomy to specific handoffs: checking work, deciding when work is done, starting scheduled work, or acting from events without a live prompt.
Are these loops ready for every task?
No. The dispatch repeats Anthropic’s caution that not every task needs a loop. The safest starting point is a bounded task with clear verification, measurable goals, and cost limits.
What remains unknown?
The source material does not confirm usage numbers, broad availability for every feature, or reliability across high-volume deployments. Some features are described as research previews.
Source: Thorsten Meyer AI
