Two years ago my inbox had a rhythm to it. “Can you build a script that pulls this report?” “Can you connect our form to our spreadsheet?” “Can you automate the thing we do by hand every Monday?” That was the business. Write the automation, hand it over, win back somebody’s afternoon.
That inbox went quiet.
Not because businesses stopped needing the work. Because the code got cheap. A tool that costs twenty dollars a month will now write a working Apps Script in about the time it takes to describe it, and most developers are already using these tools or planning to. The thing I used to charge for, the actual generating of code, is close to free.
So if you run a small business, here is the honest question: when the machine can build it, what do you still need a person for?
The 95% that go nowhere
The tempting answer is “nothing.” Buy the AI, point it at the problem, save the salary. A lot of companies are trying exactly that, and the results are rough. MIT looked at 300 real deployments and found 95% of corporate AI pilots delivered no measurable return. Gartner expects more than 40% of the new AI “agent” projects to be canceled by 2027, mostly because the cost and the value never met.
Sit with that. The models are genuinely good, and 95% of the projects still went nowhere. That gap is where the human work went.
Here is what I have actually watched happen, building these systems for clients.
Somebody has to own the thing
AI gives you an output and then it is gone. It does not get the call at 4pm when the automation chokes on a file it has never seen. It does not answer for the report that went out wrong. A business is not paying for code. It is paying for a person who owns the system, fixes it when it breaks, and carries the weight when something goes sideways. “Someone to call” is most of the value.
Somebody has to figure out what to build
Owners do not hand you a clean spec. They hand you symptoms. “Things feel chaotic.” “We keep dropping balls.” They contradict themselves, because the real process lives in three people’s heads and a spreadsheet nobody trusts. AI is excellent once the problem is written down clearly. Writing it down clearly, by sitting with a confused human and watching how they really work, is the job.
Somebody has to make it survive your actual business
Models write clean code for clean situations. Your business is not a clean situation. You have a legacy system from 2014, an API that goes down on Tuesdays, an employee who will quietly refuse to use anything new, and five edge cases nobody mentioned because they forgot they existed. Getting an automation to work, and keep working, inside one specific messy company is a different kind of labor than generating it.
Somebody has to remember why
Why it was built this way. Which client gets handled differently. Who should never be given access to the dashboard. That lives in a relationship that has been around a while, not in a fresh prompt.
The autopilot
Think about how a plane flies. The autopilot does most of the actual flying, and on a long boring stretch it does it better than a person would. We still put two trained pilots up front, and we still pay them well. Not because they out-fly the computer on a calm day. Because when the weather turns, or a sensor lies, or something happens the system was never trained for, a person has to be there who understands the whole picture and is accountable for everyone on board. The automation handles the routine. The human handles the exceptions and owns the outcome. Nobody looks at a cockpit and says the autopilot is good now, let’s lose the pilots.
Your business systems are the same. The AI is the autopilot. It is real, it is good, and it is not the one you trust with the plane.
The honest part
I am not going to oversell this. None of these are permanent walls. The models keep getting better at the messy parts too. The real risk was never that AI does my job. It is that AI lets my client, or a cheaper competitor, do my job. The edge is not “human beats machine.” It is being the specific, trusted human who uses the machine well, knows your business cold, and stands behind the result. The competitor to watch is not the AI. It is another person wielding AI better than you.
Put one person in charge
So here is the practical move, and it is smaller than it sounds. Put one person in charge of systems and automation.
One. Whether that is someone on your payroll or an outside relationship you keep does not matter much. What matters is that a single person wakes up responsible for looking across the business and asking the boring, valuable questions. What are we still doing by hand that we shouldn’t be? What did we buy and never use? What broke last week, and why? What would save the most hours if we built it right?
The MIT numbers back this up. The projects that worked were not the ones with the fanciest tools. They were the ones run through a real owner or a real partnership. Going through specialized partners succeeded about two-thirds of the time. Dabbling internally with no clear owner failed roughly three times as often.
Most businesses already have ten hours a week and a real slice of revenue trapped in manual work and half-used software. The tool to fix it is now cheap. The person to own it is the part that pays for itself.
AI took the part that was always going to get automated, the typing of the code. It left the harder part right where it has always been. Knowing what to build, making it work in the real world, and being someone a business can trust to own the result. That part you cannot prompt. It is worth hiring for, or partnering for, on purpose.