Current Agentic AI is... Workflow Automation

Wherever you stand on the OpenClaw/agentic hype, whether you’ve tried it or not, you must have at least HEARD of it.

And if you have heard of it you’ve probably seen everyone building complex “AI-native” companies where CEO agents talk to Marketing, Recruiting and Dev agents to orchestrate… Something.

Jack Dorsey’s essay specifically has been the subject of many interpretations and replication attempts, large and small.

But if you, dare I say, think this through rationally, can you REALLY trust today’s agents with running a business and making autonomous decisions? What ARE these decisions, anyway?

Lets simulate this.

A CEO agent goes over top level data (ran via predetermined cron jobs)
Said agent sets company goals in the form of metrics of some kind
The various “department agents” pick up the slack. HR agents screen CVs. Marketing agents run campaigns and monitor your SEO. Sales agents do outreach. Dev agents ship code. Support agents answer tickets.

As you can imagine, put a team of OpenClaw agents into such a configuration and you’ll have a boatload of daily reports, initiatives, ideas, and code.

More deliverables than you can reasonably review.

Perhaps you’ve also added some form of “decision points” for the agents so that they actually ping you with questions when they’re not sure, and now the agents are constantly bombarding their handlers with sometimes smart, sometimes dumb questions.

To me this sounds like glorified N8N/Make/Zapier automations.

In other words, somebody has to sit down and think this through, formalize the business practices of the company and implement them, rather rigidly, within a system of agents.

Yes, there are SOME benefits of using agents vs N8N automations:

Agents are flexible. You can ask them to change their prompts when you observe a behavior that you don’t like. On the other hand, perhaps they’re too flexible, and by fixing things ad-hoc you introduce other problems (if you don’t have strict eval processes in place)
Agents can, on occasions, do more than static workflow automations. But this can also backfire, in cases where “creative” decision making causes a catastrophe

I think that my main point is that we’re trying to replace humans with agents, but today’s agents lack critical components for this replacement to work:

Common sense
Mental models of things (not in-context “memory”, actual trained-in-knowledge)
Human cross-domain creativity
Taste
Judgment under ambiguity

So you end up with a slop-producing mammoth that still needs humans to constantly orchestrate it because we cannot trust them to make good choices, enough of the time.

This could change with the upcoming crop of models (rumored to release next week). This is the usual pattern, too. Model “X” pioneers new capabilities, model “x+1” masters them. But even if the new models are 30% smarter, what does that even mean? Does it mean they’ll be able to make good, creative and independent decisions outside of “deterministic” domains like coding?

Personally I doubt it. Hence I fail to see the logic of building grand agentic architectures to run companies.

But does this mean building such systems is a bad idea?

Not at all.

It’s actually a great idea to organize all of your organizational knowledge, including the inputs and outputs of your business, in a way that is easily accessible to AI agents. If you are able to overcome the technical hurdles and security concerns, making your business data “visible” to AI agents will:

Benefit you later, when agents DO become smart enough to operate on their own. If that ever happens (I have my doubts as you can see)
Help you work NOW. If you can do anything and everything in your business from a Claude Code or OpenClaw conversation, you’re already ahead of the curve and moving fast. I think this is the best goal you can set for yourself and your business. Not a business that “runs on its own”, but a business that YOU and your employees can run from AI terminals. And don’t fire your humans. Teach them to use AI systems instead.