@MinLiBuilds: Workflow has 6 forms. I found the original prompt and placed it in the comments. The six forms are: Classify and Execute: a classifier agent determines the task type, routing to different agents or behaviors accordingly; also can determine the output classification upon task completion. …
Summary
Introduces six forms of AI agent workflows: classify and execute, fan-out and consolidate, adversarial validation, generate and filter, tournament, loop until done.
View Cached Full Text
Cached at: 06/05/26, 05:18 PM
Workflow has 6 forms.
I found the original prompt in the comments section.
The six forms are:
Classify and Execute: A classifier agent determines the task type, routing it to different agents or behaviors accordingly; it can also determine the output’s category when the task is complete.
Fan-out and Combine: Break the task into many smaller steps, run each step with an agent, then combine all results. This pattern is especially useful when many steps can benefit from clean, non-interfering context windows. The combine step acts as a barrier: it waits for all fanned-out agents to finish, then merges their structured outputs.
Adversarial Verification: For every agent’s output, run a separate agent that adversarially verifies the output against a rubric or set of criteria.
Generate and Filter: Generate multiple ideas on a topic, then filter them based on a rubric or verification, deduplicate, and keep only the highest quality, validated ones.
Tournament: Let N agents compete against each other using different methods on the same task, rather than dividing the work among them. Paired judging agents determine the winner, proceeding until only one remains.
Loop Until Completion: For tasks with unknown workload, keep generating agents until a stopping condition is met (e.g., no new findings, no errors in logs), rather than using a fixed number of iterations.
Similar Articles
@elliotchen100: On the surface, Claude Code can generate its own workflow; looking deeper, the control plane of agent products is changing. In the past, we stuffed complex tasks into a long context, expecting the model to remember the goal, break down steps, and judge completion; now these things are starting to be externalized into executable har…
The article points out that behind Claude Code's ability to automatically generate workflows, it reflects that the control plane of AI agent products is shifting from relying on long contexts to remember goals and decompose steps, towards externalizing into an executable harness, including task structure, permission boundaries, verification mechanisms, and stop conditions.
@cellinlab: https://x.com/cellinlab/status/2064144608242679822
This article introduces the concept of Loop Engineering — instead of directly writing prompts for AI agents, it designs a system (loop) that recursively lets the agent iterate on tasks until completion. The article provides a detailed comparison of how Claude Code and Codex implement five building blocks: automations, worktrees, skills, sub-agents, etc. It suggests this could be the future trend of collaborating with coding agents, but also warns about token costs and AI slop issues.
This article systematically reviews AI Agent architecture and engineering practices, covering control flow, context engineering, tool design, memory, multi-agent organization, evaluation, tracing, and security. It is based on the OpenClaw implementation and emphasizes the critical role of Harness (testing and validation infrastructure) for system stability.
This article systematically reviews AI Agent architecture and engineering practices, covering control flow, context engineering, tool design, memory, multi-agent organization, evaluation, tracing, and security. It is based on the OpenClaw implementation and emphasizes the critical role of Harness (testing and validation infrastructure) for system stability.
@cxjwin: Found a very practical website collecting various ready-made Agent Loop templates. Core idea: stop manually prompting all the time; design the loop so the agent runs the task on its own. It includes Test Until Green, Fix CI Until G…
Shared a practical website that collects ready-made Agent Loop templates. The core idea is to let the AI agent automatically complete tasks. It includes common loops like Test Until Green, and is suitable for use with Claude Code and Cursor.
@WWTLitee: Another multi-agent collaboration tool: agency-agents. It can run separate roles like frontend, community, creative checks, and reality checks, each with its own boundaries, pace, and delivery habits, making it feel more like a team working together. The repo now has 103.5k stars, ...
Agency-Agents is an open-source AI multi-agent collaboration tool that integrates multiple specialized roles (e.g., frontend, community, creative checks) with independent boundaries and delivery habits, suitable for building multi-agent workflows. The GitHub repo has 103.5k stars.