TL;DR
Anthropic’s Claude Code team published guidance on agentic loops on June 30, 2026. Thorsten Meyer AI’s July 1 dispatch reframes the guidance as four levels of delegation, from self-checking skills to event-driven workflows.
Anthropic’s Claude Code team published guidance on agentic loops on June 30, 2026, defining a loop as an agent repeating work until a stop condition is met. A July 1 Thorsten Meyer AI dispatch reframed that guidance as a delegation ladder for teams deciding how much checking, stopping, triggering, and prompting to hand to AI systems.
The confirmed source is Anthropic’s Claude blog post, “Getting started with loops”, by Delba de Oliveira and Michael Segner. According to the source material, Anthropic describes several loop patterns in Claude Code, including turn-based skills, goal-based work, time-based loops, and proactive workflows.
Thorsten Meyer AI’s dispatch says the useful lens is not the mechanics of a loop, but what the user stops doing. In that framing, the first rung hands off the check, the second hands off the stop condition, the third hands off the trigger, and the fourth hands off the initial prompt itself.
The source material is clear about attribution: definitions and examples come from Anthropic, while the delegation ladder framing comes from the dispatch author. It also says some features are research previews, which means availability and behavior may vary as Anthropic updates Claude Code.
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.”
How Teams Hand Off Work
The framework matters because it gives developers and business teams a practical way to decide when AI should act as a tool they operate and when it can become a process that runs. That distinction affects quality control, cost, and who remains responsible for the final result.
For technical teams, the strongest advice in the source material is to start with clear checks and bounded tasks. Anthropic’s examples point toward measurable success criteria, such as a passing test suite or a performance score crossing a set threshold, rather than vague goals that leave too much room for an agent to stop early.
For managers, the ladder turns agent adoption into a question of bottlenecks. If a person is spending time rechecking repeat work, a skill-based loop may be enough. If the work arrives on a schedule, a time-based loop may fit. If work starts from external events, proactive workflows may be the higher-risk, higher-autonomy option.
AI workflow automation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Claude Code Loop Patterns
The first rung is turn-based: the user prompts the agent, and the agent works until it judges the task complete. The dispatch says the practical improvement is to encode verification in a Skill, so the agent checks its own work before returning it.
The second rung is goal-based, using a command such as /goal. In the source material’s description, the user defines a success condition and a turn cap, while an evaluator model checks whether the goal has been met before the agent stops.
The third rung is time-based, using tools such as /loop or /schedule, where a clock starts the work. The fourth is proactive, where event-driven workflows and auto mode can start work without a human prompt in real time.
“Four loop types = four rungs of delegation.”
— Thorsten Meyer AI dispatch
agentic loop automation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Preview Features And Open Limits
It is not yet clear from the source material how widely every Claude Code loop feature is available, how stable research-preview behavior will be, or how pricing will change as teams run more autonomous work. The dispatch warns that autonomy is metered, but does not provide independent cost data.
The practical gains are also not independently measured in the source material. Claims about saved time, better verification, or larger agent runs should be treated as implementation guidance rather than proven business outcomes until teams test them in their own codebases and workflows.
AI task delegation platforms
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Pilot Criteria Before Larger Runs
The next step for teams is to test the ladder on small, measurable tasks: write the check, define the goal, set a turn cap, and monitor usage before moving to scheduled or proactive work. The source material recommends choosing the right primitive, using the cheapest capable model, and fixing the system rather than only repairing one failed run.
Anthropic’s Claude Code documentation at code.claude.com/docs is the place to watch for feature changes. For now, the safest reading of the dispatch is incremental: hand off one piece of work at a time, then move up the ladder only when the task has earned more autonomy.
AI process management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the actual news here?
Anthropic published Claude Code guidance on agentic loops on June 30, 2026. Thorsten Meyer AI then framed those loop types as a four-rung delegation ladder on July 1.
What are the four loop types?
The source material lists turn-based skills, goal-based loops, time-based loops, and proactive workflows. Each one gives the agent more responsibility for how work is checked, stopped, started, or initiated.
Is the delegation ladder Anthropic’s term?
No. The source material says Anthropic supplied the definitions and examples, while the delegation ladder is the dispatch author’s framing.
When should teams use goal-based loops?
They fit tasks with a measurable success condition, such as tests passing or a score crossing a threshold. Vague goals are a weaker fit because the agent may judge the work complete too early.
What is still unknown?
The source material leaves open the availability of preview features, real-world cost at scale, and the reliability of larger autonomous workflows. Teams will need pilots before broad rollout.
Source: Thorsten Meyer AI
