Kayode Faturoti is an internet entrepreneur and co-founder of Liners, Breet, Cardtonic, Stack Directory, and Homevy.
getty
"My company is run by AI agents." It is a great line to drop online. It sounds like the future showed up early and handed you the keys.
At Liners, the software directory I run, nine AI agents do most of the work: research, content, quality checks, review moderation and even shipping code. So I can tell you the line can be true. I can also tell you it is nowhere near as effortless as it sounds.
I built our platform with agents from the ground up, and not for the novelty of it. A directory has too many moving parts to run by hand at any real scale. You have to discover products, confirm they are real, pull their details from across the internet, watch them for changes, moderate user reviews, publish and notify the right people. One agent researches and enriches data. Another turns that research into copy. One checks listings before they go live. Another moderates reviews. Engineering agents implement and debug. Spread across the workflow, they let a small operation cover ground that would normally need a crowd.
That is the part that lives up to the hype. And here are the parts that do not:
1. Agents need far more instructions than people, and you never finish writing them.
A human employee arrives with common sense. Tell a new hire to "use your judgment," and they mostly will, because they have lived in the world and can reason from it. An agent has not and cannot. So, every task turns into a long list of dos and don'ts, with edge cases spelled out and exceptions named one by one. Remember, you are not writing a job description; you are writing instructions for a worker with no instinct, and the level of detail required is far beyond anything you would give a real person.
You also never get to stop. The moment your environment shifts, with a new product category, a new flavor of spam review or a new way users try to game the system, the instructions need another pass. The training is not setup work you do once; the training is the job.
2. The more agents run in the background, the harder they are to see.
A handful of agents quietly working while you sleep sounds like pure leverage. It is, right up until something goes sideways and you have no idea which agent did what, or when. Oversight does not scale on its own.
The fix that worked for me was almost embarrassingly simple. I made every agent report what it does, in plain language, into one Slack channel. It's a running "tell me everything you did" feed. Later, I made a version of that feed public on the site. If a machine is acting on your behalf, you should be able to read back exactly what it did, in order, without going digging. Without that log, you are not really running an agent; you are hoping.
3. Every clever move costs money.
Agents run on tokens, and tokens are cash. A teammate's salary is the same whether their day is busy or slow. An agent's bill moves with every task you let it perform, and "let's have it double-check everything twice" sounds good until the invoice arrives. If left unwatched, the cost climbs quickly.
Two things helped me with this. Newer models follow instructions noticeably better, so you waste fewer retries cleaning up after them. And keeping the temperature low keeps output predictable instead of creative when you did not ask for creative. To sum it up, predictable is cheaper.
4. AI agents are brilliant at some things and untrustworthy at others.
Agents are excellent at anything close to code and structured work, and weak anywhere real judgment is required. They will implement a feature or extract data all day without complaint. But ask them to decide whether a borderline review is genuine or a vendor is quietly gaming the system, and they wobble.
That is why quality assurance never goes away. The agents do most of the work, but a human still has to own the call on anything subjective, reputational or hard to undo.
So, does running your operations on AI agents actually work?
Yes, but treat the phrase with suspicion whenever you hear it, including from me.
Most versions of that claim are harder behind the scenes than they look from the outside. The people making it work are not the ones who flipped a switch and walked away. They are the ones writing relentless instructions, logging everything, watching the meter and keeping a human hand on every decision that matters.
If you are about to hand your operations to agents, go in with your eyes open. They do not remove the work so much as change its shape: less doing and more directing, reviewing and correcting. If it’s done lazily, it becomes a mess that scales. But if done well, it is the most leverage a small team has ever had. Just do not expect it to feel as simple as it sounds.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?