AI agents
Agentic workflow
An agentic workflow is a multi-step process in which a language model plans or performs some of the steps with tools, ranging from fixed code paths that call a model at set points to open-ended agents that choose every step.
What it is
"Agentic workflow" covers a spectrum. At one end is a fixed sequence, written in code, that calls a model at chosen points: classify this email, then route it. At the other end is an AI agent that decides every step itself. Anthropic's engineering guide calls the first kind workflows and the second agents, and groups both as agentic systems. It describes several workflow patterns, among them parallelisation, where independent pieces run at once and their results are combined.
Why it matters
The further toward full autonomy a process sits, the more it can handle the unexpected, and the more it costs and the harder it is to predict. The practical advice from Anthropic's guide is to start with the simplest approach that works and add autonomy only when the task needs it. Most business processes are well defined once someone has done them a few times; they do not need a model to re-decide every step on every run.
How it works
A typical progression:
- A person asks an agent to do a task once, interactively.
- The agent works out the steps, calling tools and reading results.
- The steps that worked are saved as a fixed procedure.
- The procedure runs again on demand or on a schedule, with the model used only where judgement is needed.
Example
An accounts team asks an agent each Monday for clients with unbilled support hours. The first time, the agent explores: which PSA fields hold billable time, how customers are named in the accounting system. After that, the same three fetches and one match run as a saved procedure every Monday at 8am, and the agent is asked only to write the summary.
Common pitfalls
- Autonomy where a script would do. Paying a model to rediscover the same steps every run costs tokens and adds variance.
- No checkpoints. Long chains with no human checkpoint let one early mistake flow through every later step.
- Triggers nobody can see. Event-driven chains spread across many tools are hard to audit; you cannot easily answer what started a run and why.
How actionlake handles it
In actionlake, a task an agent has done once becomes an agent plan: a saved, versioned procedure of actions, cross-system joins, loops, waits and approval steps. Plans run on demand or on a schedule. By design there are no event triggers, webhooks or polling triggers: agents and schedules start the work, and a scheduled plan can ask "what is new since the last run?" to cover most of those cases. Every run is recorded with its intent in the ledger. Hand your agent your written procedure and it builds the plan. See the platform page.
Questions
Is an agentic workflow the same as an AI agent?
Not always. An agentic workflow can be a fixed sequence that calls a model at set points, or it can be a fully autonomous agent. Agents are the most open-ended kind.
When should I use a fixed workflow instead of an agent?
When the steps are known and repeat. Save them as a procedure and let a model handle only the parts that need judgement, such as writing a summary.
Can an agentic workflow run without a person?
It can run on a schedule unattended, but high-impact steps should still pause for a named person's approval, and every run should be recorded.
Related terms
- AI agentAn AI agent is software in which a language model decides its own next steps and tool calls in a loop, using results from its environment, until a goal is met or a stopping condition is reached.
- Agent planAn agent plan is a saved, versioned procedure of actions, cross-system joins, loops, waits and approval steps that an AI agent writes once and that then runs the same way on demand or on a schedule, without the model re-deciding every step.
- Human-in-the-loop approvalHuman-in-the-loop approval is a control that pauses an AI agent's action until a named person approves or rejects it, so high-impact or irreversible actions never run on the model's judgement alone.
- Cross-system joinA cross-system join combines records from two or more separate systems that refer to the same real-world thing, such as one customer in an accounting ledger and a ticket desk, and returns a single combined answer.
- Tool callingTool calling, also called function calling, is a language model's ability to answer with a structured request to run a named function with arguments, which the application executes before handing the result back to the model.
Sources
Connect an agent. Start free.
100,000 calls free every month. No card. Hard caps on every key.