Your 'AI Agent' Is Probably Just a Cron Job

The short version

Most things marketed as AI agents are actually automations or workflows with a chatbot bolted on. The distinction matters because it changes your architecture, pricing, reliability story, and customer promise. Automations follow rules. Workflows add a language model at fixed steps in a deterministic pipeline. Agents decide what to do next based on context, adapt to unexpected input, and produce variable outputs. Most enterprise AI value today lives in the workflow tier. Agents get the headlines, but well-built AI workflows get the work done. Don't oversell a cron job as an agent because the word sounds better in the pitch deck.

Everyone's shipping "AI agents" right now. Most of them aren't agents. They're automations or workflows wearing a trench coat.

This isn't just a naming problem. If you're building product around AI, the distinction between an automation, a workflow, and a real agent changes your architecture, your pricing model, your reliability story, and what you promise customers. Get it wrong and you set expectations you can't meet.

I run 18 agents as part of my PM practice. Some of them are genuinely agentic. Some are honestly closer to fancy automations. Knowing which is which helps me set the right expectations and invest tuning time where it matters.

Three things that look the same but aren't

Automations follow rules. If X happens, do Y. No judgment, no variation. Your email drip campaign that sends a message on day 3 after signup is an automation. It's fast, reliable, and does exactly what you told it to. It'll also do exactly the wrong thing if the conditions change and nobody updates the rules.

AI workflows add a language model to a deterministic pipeline. You get a support ticket, an LLM classifies it, routes it to the right queue, maybe drafts a response. There's intelligence at specific steps, but the overall flow is still fixed. Step 1 always leads to step 2. The AI makes individual steps smarter, but it doesn't decide what steps to take.

Real agents make decisions about what to do next based on context. They can change their approach when something unexpected happens. They compose different skills together depending on the situation. They're non-deterministic, which is both their power and their risk.

How to tell what you're actually looking at

Ask these questions about any "AI agent" you're evaluating or building:

Does it decide what to do, or just how to do a predefined thing? If the sequence of steps is hardcoded and the AI only handles execution within each step, it's a workflow. If the AI decides which steps to take based on what it's seeing, it's closer to an agent.

Can it handle a situation it wasn't explicitly designed for? Automations break on edge cases. Workflows handle them poorly. Agents can (sometimes) adapt. If your "agent" falls over the moment something unexpected happens, it's not an agent.

Does it learn from feedback? True agents adjust their behavior over time. Most things marketed as agents today don't. They run the same way every time, which is fine, but call it what it is.

Is the output predictable? If you run it ten times on the same input and get the same result every time, it's an automation or a tightly constrained workflow. Agents produce variable outputs because they're making judgment calls.

Why this matters for product work

If you're a PM building AI features, this distinction shapes your entire approach:

Reliability expectations are different. Automations should work 100% of the time. Workflows should be highly reliable with occasional AI missteps. Agents are inherently less predictable. You need to design the right guardrails and human-in-the-loop checkpoints for each type.

Pricing models change. Automations work great on per-seat or per-action pricing. Workflows can be priced by volume. Agents, if they're truly autonomous, should be priced on outcomes because that's what you're selling. If you price an agent like a workflow, you're either leaving money on the table or setting yourself up for disappointed customers.

The build vs. buy decision shifts. Automations are commoditized. Workflows are becoming commoditized. True agentic capabilities are still hard to build well. That's where the moat is, if you can get the reliability right.

What I've learned from running my own

My daily focus agent is honestly more of an AI workflow. It pulls from fixed data sources, runs a prompt, and gives me a summary. The sequence doesn't change. The AI just makes the output useful. And that's fine. I don't need it to be agentic. I need it to work every morning.

My competitive intel agent, on the other hand, is more agentic. It decides which competitors to dig into based on what changed, chooses which sources to prioritize, and adjusts its depth depending on what it finds. It surprises me sometimes, which is the hallmark of something genuinely agentic.

The point isn't that agents are better than workflows or automations. Each has its place. The point is knowing which one you need for a given job, and not overselling a workflow as an agent just because the word sounds better in a pitch deck.

The honest taxonomy

When I set up a new automated process, I ask myself:

Is this a rule? Build an automation. Fast, cheap, reliable.

Is this a fixed process where some steps need intelligence? Build a workflow with AI. Predictable structure, smart execution.

Is this a problem where the right approach depends on what the AI discovers along the way? Build an agent. Accept the unpredictability. Invest in guardrails.

Most of the real value in enterprise AI today lives in the workflow tier. Agents get the headlines, but well-built AI workflows get the work done. Don't let the hype push you past what your use case actually needs.