TL;DR
Anthropic’s Claude Code team published a June 30, 2026 framework defining agentic loops as repeated work cycles that run until a stop condition is met. Thorsten Meyer AI’s July 1 dispatch reframes the four loop types as a delegation ladder, showing what users hand off at each step.
Anthropic’s Claude Code team has published a new framework for agentic loops, defining them as repeated cycles of work that continue until a stop condition is met, a development that matters for teams deciding how much AI work to delegate and how much human control to keep.
The framework, cited in a July 1, 2026 Thorsten Meyer AI dispatch, identifies four loop types: turn-based, goal-based, time-based and proactive workflows. Thorsten Meyer AI frames those categories as a delegation ladder, where each rung represents one more part of the work that the user no longer performs directly.
In the turn-based loop, the user still starts each task, but the agent can be given a skill that checks its own output. In the goal-based loop, the user defines the desired result and an evaluator model sends the agent back to work until that result is met or a cap is reached.
The higher rungs shift more control away from real-time human prompting. A time-based loop starts work on an interval, while proactive workflows can be event-driven and may coordinate multiple agents. The source material says some features are research previews, so availability may vary by user and environment.
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.”
Delegation Becomes a Design Choice
The practical point is that teams are no longer only choosing prompts; they are choosing where human judgment sits in the workflow. A loop can remove the need to repeat checks, restart scheduled work, or decide when a task is complete, but each step also increases the need for clear limits.
For developers, the framework offers a way to match task type to control level. For business readers, the same model translates into an operations question: which recurring task is the bottleneck, and which single part can be handed off without losing oversight?
The dispatch also stresses cost and quality controls. It recommends the right primitive, the cheapest capable model, clear stop criteria, pilots before large runs and watching usage, since autonomous work is metered and can become expensive if goals are vague.
Agentic AI Engineering: Building AI Agents for Beginners: A Hands-On Guide to No-Code Workflows, LLM Tools, RAG, Automation, and Safe Multi-Agent Systems
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Prompting to Process Design
The Anthropic post, titled “Getting started with loops” and attributed to Delba de Oliveira and Michael Segner, was published on the Claude blog on June 30, 2026. Thorsten Meyer AI’s dispatch says the definitions, primitives and examples come from Anthropic, while the delegation ladder framing is its own interpretation.
The dispatch’s first rung, turn-based skills, describes a familiar workflow: a user prompts an agent, the agent edits or researches, checks its work and returns a result. Anthropic’s example, according to the source material, involves a front-end skill that does not mark a UI change done until it starts the dev server, clicks the control, captures screenshots, checks the console and runs a performance trace.
The second rung, goal-based work, is presented as useful when iteration matters. The dispatch gives the example of asking an agent to raise a homepage performance score above 90 while limiting the run to five attempts, making the turn cap a cost control.
“Strip the hype and a loop is simple: an agent repeating work until a stop condition is met.”
— Thorsten Meyer AI dispatch
Designing Multi-Agent Systems: Principles, Patterns, and Implementation for AI Agents
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limits Remain Product-Dependent
It is not yet clear from the provided material how broadly each loop type is available across Claude Code users, paid plans or deployment settings. The source says some features are research previews, which means teams may not be able to use every pattern in production today.
The dispatch also does not provide measured results showing how much time, cost or error rate changes when teams move from manual prompting to goal-based or proactive workflows. The business value remains a planning claim unless teams test it against their own workflows.
Digital Transformation: How to Lead Business Transformation through Digital Platforms and AI
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Teams Test Their First Rung
The next step for readers is to identify one recurring task and choose the lowest useful rung: self-checking skills for quality control, goal-based loops for measurable outcomes, scheduled loops for recurring work or proactive workflows for event-driven operations.
Anthropic’s docs at code.claude.com/docs are the expected reference point for implementation details, while the dispatch’s guidance points teams toward small pilots before running many agents or moving business processes into higher-autonomy workflows.
AI Automation for Businesses: Practical AI Strategies for Modern Small and Medium Businesses
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Anthropic publish?
Anthropic’s Claude Code team published a framework describing agentic loops as repeated work cycles that continue until a defined stop condition is met.
What is the delegation ladder?
The delegation ladder is Thorsten Meyer AI’s framing of Anthropic’s loop types. It describes how users hand off more work at each level: the check, the stop condition, the trigger and finally the prompt.
Which loop should a team start with?
The source material recommends starting with the simplest working pattern. For many teams, that means a turn-based skill that verifies its own work before moving to scheduled or proactive systems.
What are the main risks?
The main risks are unclear goals, weak verification, runaway costs and premature autonomy. The dispatch recommends clear stop criteria, turn caps, small pilots and usage monitoring.
Are all four loop types fully available now?
The provided source says some features are research previews. Availability may depend on the user’s tools, environment and Claude Code access.
Source: Thorsten Meyer AI
