If a Company Can't Afford Humans, It Can't Afford AI
Last week, Jack Dorsey cut 4,000 people from his company, Block. The reason: “intelligence tools.” Block’s stock jumped 25%. A former employee called it “organizational bloat wearing an AI costume.” Even Sam Altman admitted there is “AI washing” happening across the industry.
I have spent 25 years building software. I have designed automation systems, built integrations, and watched every wave of “this changes everything” technology arrive with promises and leave behind lessons. I am not anti-AI. I use it daily. I build with it. I advise companies on how to implement it.
But I need to say something that the current AI conversation desperately needs to hear.
Most companies laying off workers in the name of AI are not replacing those workers with AI. They are just laying off workers.
The Convenient Narrative
AI is the best excuse corporate leadership has had in decades. It sounds forward-thinking. It sounds inevitable. It sounds like strategy. It is a perfect presentation for the corrections from pandemic-era overhiring dressed up as technological progress. The Agentic era arrives just-in-time to improve the business bottom-line.
But can AI agents take over the tasks in a workflow? There is a meaningful difference between “we have built systems that can now perform these tasks” and “we believe AI will eventually perform these tasks, so we are cutting headcount now.” The first is engineering. The second is speculation dressed up as strategy.
What Agentic AI Actually Requires
Let me explain what an AI agent actually needs to run a workflow. Not in theory. In practice.
An agentic system chains together multiple AI components such as LLM, data tools, API, and decision points to execute a sequence of tasks. An orchestrator agent decides which tool to call, in what order, based on the current state of the work. It can loop, self-correct, and hand off to specialized sub-agents. The architecture is real. It works.
But let’s be clear about what agents can do today. An agent can create a slide. Add an entry to a database. Pull data from a system provided that the data connector is linked. These are tasks. Discrete, bounded, repeatable. What agents cannot do yet is develop and manage an entire system autonomously. The complexity of orchestrating a full business workflow, with its dependencies, exceptions, and human judgment, is beyond what an unmanned AI agent can handle. That gap between task and system is where most companies lose the plot.
From the outside, it looks like the agent is learning. It looks like it figured out what to do next. Users see a polished output and assume the system taught itself. That is the illusion. Behind every functioning agentic workflow is a human. Usually a team of humans. Usually with significant technical skill. They engineer the loop. Define the exit conditions. Map which agent calls which tool, what data connector each agent needs, and what happens when something goes wrong. Build error handling for every branch. Test the chain against real-world edge cases, not the clean demo data. In my experience, the flow is deterministic. It is programming logic.
None of this happens on its own. It takes a team of humans with significant technical skill. And the iteration behind the scenes to get a workflow running reliably takes weeks, sometimes months, to establish.
The agent does not figure out your workflow by watching your team. It does not absorb 15 years of institutional knowledge from a training manual. Every decision point, every conditional branch, every “well, it depends” judgment call that your experienced staff makes instinctively has to be explicitly identified, documented, and translated into logic the system can follow. Before any of that engineering begins, someone has to know what the workflow actually is. Not what the process manual says. What people actually do. The workarounds, the judgment calls, the tribal knowledge that lives in no system.
Most organizations have not done that work. Which means they are not ready for agentic AI. They are not even ready to describe what they would automate.
So when a company says “we are replacing these roles with AI,” ask one question: have you built the system that replaces them? Not “are you planning to.” Not “does your vendor say it is possible.” Have you built it? Is it running?
In most cases, the answer is no. The technology is advancing fast. The organizational readiness is not.
I believe agentic AI can run your workflow. But first, you need to actually know what your workflow is. And right now, most companies do not.
The truth is that no amount of layoffs will make AI work. Instead of letting people go, why not augment what they do? Human-AI collaboration unlocks value that never existed before. That is the real opportunity the technology offers.
.
