TL;DR
Anthropic’s Claude Code team published a plain definition of agentic loops on June 30, 2026, and Thorsten Meyer AI followed on July 1 with a four-rung delegation framework. The confirmed material identifies turn-based skills, goal-based /goal, time-based /loop or /schedule, and proactive workflows with auto mode, while source notes say some features remain research previews. The practical issue for readers is how much checking, stopping, triggering and prompting they are ready to hand to agents.
Thorsten Meyer AI published an Insights AI Dispatch on July 1, 2026 that recasts Anthropic’s new Claude Code guidance on loops as a four-rung delegation ladder, giving developers and business teams a structured way to decide how much work to hand to AI agents.
The dispatch says the base definition comes from Anthropic’s Claude blog: an agent repeats cycles of work until a stop condition is met. Thorsten Meyer AI’s contribution is the delegation ladder framing, which classifies loops by what the user stops doing: checking, deciding when to stop, starting the run or writing the prompt.
The first rung is turn-based skills, where a person still starts each turn but encodes verification into a Skill. The example cited from Anthropic is a front-end skill that checks a UI change by starting the server, clicking the control, comparing screenshots, checking the console and running a performance trace before treating the task as done.
The higher rungs add more autonomy: goal-based /goal uses an evaluator model and a turn cap, time-based /loop or /schedule starts work on an interval, and proactive workflows with auto mode can respond to events without a human prompt in real time. The source note says some of these features are research previews.
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.”
Teams Get a Delegation Map
The practical takeaway for readers is that the framework turns agentic AI from a vague label into a set of delegation choices. A team can ask which bottleneck it wants to remove: manual review, repeated prompting, scheduled starts or the need for a person to ask in the first place.
That distinction matters for cost and control. The dispatch advises using the right primitive, the cheapest capable model, clear stop criteria and pilot runs before scaling to hundreds of agents. Those are cost controls, not confirmed savings figures.
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.
Anthropic’s Four Loop Types
The source material ties the framework to Getting started with loops, a Claude blog post by Delba de Oliveira and Michael Segner published June 30, 2026. According to Thorsten Meyer AI, Anthropic supplied the definitions, primitives and examples, while the delegation ladder is the site’s own framing.
The ladder runs from manual turn-based work to autonomous proactive workflows. At the lowest rung, the user still prompts and inspects. At the highest, an event or schedule can trigger a multi-agent workflow, with each task governed by its own goal.
The dispatch also says output quality depends more on the surrounding system than the loop itself: a clean codebase, reusable Skills, fresh-context review agents and fixes to the workflow after failed runs. That guidance is presented as operating advice, not a measured benchmark.
“a loop is an agent repeating cycles of work until a stop condition is met”
— Anthropic Claude blog, Delba de Oliveira and Michael Segner
Workflow Automation with Microsoft Power Automate: Design and scale AI-powered cloud and desktop workflows using low-code automation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Preview Status and Open Costs
Several points remain unsettled. The dispatch’s source note says some features are research previews, so availability, product limits and production support may vary by account or release channel. The article does not provide independent tests of /goal, /loop, /schedule or proactive workflows.
It is also unclear how broadly teams will adopt the ladder framing, how reliable event-driven workflows will be across messy real-world tasks, or how quickly costs rise when agents iterate. The strongest confirmed point is narrower: Anthropic published loop guidance, and Thorsten Meyer AI framed it as a delegation model.
The AI Prompt Playbook: Master AI Prompt Engineering with 140 Ready-to-Use Templates for ChatGPT, Claude, Gemini & Copilot
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Pilots Before Larger Runs
The next step for readers is likely testing the ladder on bounded work. A safe starting point is a turn-based Skill with measurable checks, followed by a /goal pilot for tasks with a clear pass condition, such as tests passing or a score crossing a fixed threshold.
For scheduled or event-driven work, teams will need to watch Claude Code documentation, preview availability and usage reporting. The dispatch points readers toward code.claude.com/docs and advises watching /usage before larger runs.
The SQL Workshop: Learn to create, manipulate and secure data and manage relational databases with SQL
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the Delegation Ladder?
It is Thorsten Meyer AI’s framing of Anthropic’s loop guidance as four delegation rungs: checking, stop conditions, triggers and prompts.
Did Anthropic create the ladder framing?
No. The source note says Anthropic supplied the loop definitions, primitives and examples, while the ladder framing comes from Thorsten Meyer AI.
What are the four loop types?
The four named types are turn-based skills, goal-based /goal, time-based /loop or /schedule and proactive workflows with auto mode.
Are these features ready for production?
The source note says some features are research previews. Production readiness, limits and support are not fully detailed in the supplied material.
Why would a business reader care?
The framework helps teams decide what to delegate and what to keep under human control, especially around quality checks, costs and scheduled work.
Source: Thorsten Meyer AI
