Most things being sold to PMs as AI agents are not agents. They are automations or workflows wearing a trench coat. This is not just a naming problem. If you are building product around AI, the distinction changes your architecture, your pricing, your reliability story, and what you can promise customers.
Three things that look the same but are not
Automations follow rules. If X happens, do Y. No judgment, no variation. An email drip that sends a message on day 3 after signup is an automation. It is fast and reliable, and it will do exactly the wrong thing if conditions change and nobody updates the rules.
AI workflows add a language model to a deterministic pipeline. A support ticket comes in, an LLM classifies it, routes it, maybe drafts a reply. There is intelligence at specific steps, but the flow is fixed. Step 1 always leads to step 2. The AI makes steps smarter. It does not decide which steps to take.
Real agents decide what to do next based on context. They change approach when something unexpected happens, compose different skills depending on the situation, and produce variable outputs. That non-determinism is both their power and their risk.
How to tell what you are looking at
Ask four questions about any "agent" you are evaluating. Does it decide what to do, or only how to do a predefined thing? Can it handle a situation it was not designed for? Does it learn from feedback? Is the output predictable across runs? Run it ten times on the same input. If you get the same result every time, it is an automation or a tightly constrained workflow, not an agent.
I run a fleet of agents as part of my PM practice. Some are genuinely agentic. My daily focus agent is honestly closer to a workflow: it pulls from fixed sources, runs a prompt, gives me a summary, and the sequence never changes. My competitive intel agent is agentic: it decides which competitors to dig into based on what changed, picks sources, and adjusts depth as it goes. It surprises me sometimes, which is the hallmark.
Why it matters for the work
Reliability expectations differ. Automations should work 100% of the time. Workflows should be highly reliable with occasional AI missteps. Agents are less predictable and need guardrails and human-in-the-loop checkpoints. Pricing changes too: automations fit per-action pricing, workflows fit per-task or per-seat, and truly autonomous agents should be priced on outcomes.
Most of the real value in enterprise AI today lives in the workflow tier. Agents get the headlines, but well-built workflows get the work done. Pick one "agent" on your roadmap and run it through the four questions. If it fails one, call it what it is and build it that way.