Most AI Agents Are Just Automation in a Trench Coat (Here's the Line I Drew While Building)

Everyone's calling their product an AI agent right now.

Most of them aren't.

They're automation with a chat interface on top. Which is fine — automation is genuinely useful. But calling it an agent creates a real confusion problem, especially when you're building something that's actually trying to be one.

I ran into this while building Sendlume. And I had to get honest with myself about where the line actually is.

The difference I kept coming back to

Automation answers: If this happens, do what?
You design the workflow. It executes it.

AI agents answer: Here's the goal. Now what?
You describe the outcome. It figures out the steps.

The first is predictable. The second is adaptive.

Classic automation — Zapier, Make, n8n — is incredible for repeatable, known workflows. But it only does exactly what you programmed. Change the situation, the workflow breaks.

An agent doesn't need a pre-mapped path. You give it a goal like "research AI startups launching this month and organize what you find" and it decides what to do next at each step.

That's a fundamentally different thing.

What I'm actually unsure about

Building Sendlume — which is trying to let users create and run agents just by describing what they want in plain language — I keep running into the same hard questions:

• Most users don't know or care about this distinction. Does it matter for adoption, or only for building?
• Where does "agent" stop being a useful label and start being marketing?
• Is "describe your goal, get a workflow" actually better UX than node-based builders — or just different?
• What makes someone switch from a tool they already trust?

I don't have clean answers yet.

What Sendlume is trying to do

Instead of connecting nodes manually, you'd describe what you want:

"Find SaaS founders who posted about hiring frustrations this week, qualify them, and draft outreach."

The system generates and runs the workflow. No manual wiring.

Whether that's genuinely better or just a different interface — that's what I'm trying to find out.

One thing worth mentioning

The waitlist for Sendlume is live right now at sendlume.com.

When we launch, it won't be open to the public. Access will be invite-only — we're keeping the first cohort small so we can actually work closely with early users and build this the right way.

If that sounds interesting, the waitlist is where to be. No launch date promise, no hype — just early access before the doors close.

What I'd actually like to know

If you've built with agents or tried to explain them to non-technical users:

• Where does the automation vs agent framing break down in practice?
• Have you seen users actually prefer goal-based input over node builders — or do they revert to wanting control?
• What made you trust a new workflow tool enough to put real work through it?

Not looking for validation. Looking for where this thinking is wrong.