We ship our own products
"Agent" and "agentic" get used as if they mean the same thing. They don't, and the difference decides how much you should build.
The short answer
An AI agent is a model that can take actions: call a tool, look something up, send a message. An agentic system plans those actions, runs them in a loop, checks the result and decides what to do next, until a goal is met.
An agent can act. An agentic system can keep going, and notice when it's wrong.
What an agent is
Give a model a set of tools and a task, and it chooses which tool to call with which input. Most useful assistants are agents in this sense: they look up an order, book a slot, open a ticket.
tools = [order_status, send_whatsapp]
reply = model.run("Where is order #4821?", tools)
What makes it agentic
The loop. The system breaks a goal into steps, runs a step, reads what came back, and plans the next one. It needs memory, limits and a way to stop.
- Planning turns a goal into steps it can check.
- Reflection reads results and corrects course.
- Guardrails means budgets, permissions and a human hand-off.
When to build which
If the task is one or two steps and the right tool is obvious, a single agent is simpler, cheaper and easier to trust. Reach for an agentic loop when the path isn't known in advance: research, multi-system workflows, long-running jobs.
Getting it to production
Treat it like any production system: evaluation sets before launch, tracing for every step, cost and latency budgets, and a clear hand-off to a person. That's the difference between a demo and something your team relies on.
Written by the SanctumCloud team
Talk to us Engineers and AI builders who ship assistants and agents into real products.