Front page / Concepts / Workflows, not agents
Workflows, not agents
A workflow is a fixed sequence of steps where the model fills in each step: classify the ticket, draft the reply, check the tone, file it. The path is decided by you in advance, not by the model at run time. For recurring work with a known shape, a workflow is cheaper, faster and far more predictable than an agent deciding what to do.
Where it stands
Among the 14 concepts. Bars: GitHub stars gained this week.
- MCP and tool use+3.9k
- Agentic loops+1.3k
- Knowledge graphs (GraphRAG)+1.2k
- Retrieval (RAG)+1.2k
- Prompt engineering+1.1k
- Vibe coding+980
- Workflows, not agents+565
- Spec-driven work+16k
- Memory+7.3k
- Agent skills+5.7k
- Multi-agent orchestration+5.6k
- Computer use+4.7k
- Context engineering+3.3k
- Evals and verification+929
How to use it today
In ChatGPT or Claude, as a person doing the work, not as a developer.
- 1Write the steps of a recurring task as a numbered list with the input and output of each step; if you can, it is a workflow.
- 2Run it as a sequence of separate asks, each with its own instructions, rather than one big request; or install a skill that encodes the sequence.
- 3Put a check between steps: a format the next step requires, a rule the output must satisfy.
- 4Automate it only once it has run by hand a few times without edits.
- 5Reserve agent mode for the steps that genuinely need judgment about what to do next.
What changes day to day
Recurring work stops being a conversation and becomes a form you run. You spend time designing the sequence once, then only on the exceptions it flags.
Where it goes wrong
Forcing judgment-heavy work into a fixed sequence produces confident nonsense at the step that needed a decision. Workflows rot when the underlying process changes and nobody updates the steps.
Listings that use it well
Chosen by hand after reading them, each with the reason. Install from the listing's own page.
| Listing | Kind | Why it is here | Stars |
|---|---|---|---|
| engineering | Agent skills | workflow-builder writes deterministic workflow scripts for Claude Code's Workflow tool: fan-out, pipeline, loop, judge panel. | 29k |
| PM Skills | Agent skills | utility-pm-workflow-orchestrator runs an ordered chain of PM skills with go/no-go gates and stops on a failed step. | 791 |
| activepieces | MCP server | Open source workflow automation platform with 400 pieces exposed as MCP servers. | 25k |
| n8n-mcp | Agent skills | An MCP that lets an assistant build and validate n8n workflows node by node. | 23k |
| mcp-agent | MCP server | Implements the Building Effective Agents workflow patterns (chaining, routing, parallel, orchestrator) on MCP. | 8.6k |
Elsewhere
Repositories that embody the idea and are not in the catalog.
- langchain-ai/langgraph ↗Explicit state graphs: nodes, edges and a shared state object for agent workflows.
- n8n-io/n8n ↗Visual workflow automation with AI nodes; the predefined-path end of the spectrum.
- langgenius/dify ↗Open LLM app platform combining workflow builder, RAG pipeline and agents.
Standing
Our judgment, with the reason and the sources it rests on.
Established since 2024-12. The workflows versus agents distinction became the standard design advice in late 2024 and both main vendors still give it. Anthropic's December 2024 guide drew the line between workflows (predefined code paths) and agents (model-directed), and OpenAI's April 2025 guide gives the same advice: start simple, add agency only when needed. Settled practice: the methods are known, the tools are mature, and the question is how well you use it.
- Building Effective AI Agents2024-12-19
- A practical guide to building agents2025-04-17
Related: Agentic loops, Agent skills, Evals and verification