The Nostradamus Trap
Build, Buy, Wait. Or All Three?
If we could nudge Nostradamus and ask him for one last prophecy, where AI will be in five years, he would, as was his style, make it so vague that, in hindsight, any outcome would look prophetic. Kurzweil prefers exponential curves with dates attached. Harari warns that we may not be headed to a good place. Makers of AI swing from utopian futures to end of days in the same interview. AI prophecies emerge from the woodwork every day.
None of them is very helpful when you are running a $100 million manufacturing business and trying to decide what, if anything, to do about AI next week.
Questions you have already answered, but still wrestle with:
Our systems are old, our data messy, half our processes depend on Bob. Can AI fix that?
ChatGPT makes my holiday reservations and shops online for me. Can I set my business to similar levels of AI autonomy? Safely? I know I can’t, but thought I’d ask.
Half the company is using AI. I suspect the other half is probably using it without telling us. Do we have the talent to go DIY? Maybe we do. But should we take on automating our business processes?
If we buy or rent instead, will it fit the business we actually have? Or the hypothetical business in the vendor’s demo app? How will it sit within our existing systems and messy data? Do they know what they are doing? How did the vendor become such an ‘AI and agentic AI’ expert in just two years?
Perhaps we should wait until the dust settles. 1-800-Nostradamus keeps going to voicemail.
Three reasonable ways to take a wrong turn
You are a mid-market manufacturer, logistics company, or industrial services business. Your experience and operating reality have inoculated you against hype, especially technology hype. And you are not technologically naïve by any means.
But AI feels different. The capabilities are extraordinary, the pace of change disorienting. Everybody is talking about it. The demos are great.
Three reasonable choices present themselves.
Build. You know your business better than anyone. AI development tools have opened up a whole new DIY world. Why not create exactly what you need and cut the tether or yoke that’s allowed technology vendors to both help and hold you hostage? Reasonable. You certainly should own the processes that differentiate you. But do you need to build the technology yourself in order to own and control those processes? Or will that inadvertently become a distraction in and of itself?
Buy. You have been buying enterprise software for decades. Let the vendors make the investments and carry the technical burden. Also reasonable. But they too are grappling with their own AI-triggered questions, stranded between the terra firma of the past and the promise and risks of a new, emerging, untested AI. Is this the right time to go all-in with an “AI-first,” whatever that means, vendor?
Wait. Let the market mature. Avoid premature commitments. Learn from other people’s expensive mistakes. Perhaps the most reasonable choice of all. Until your competitors have spent two years building muscle and experience that leaves you permanently disadvantaged.
You, your competition, the market, and vendors are all being buffeted by the same AI winds. With so many simultaneous confounding factors, it is tempting to lean on prophets and freshly minted AI futurists rather than on the grounded, controlled judgment that has served you well for years. That is the Nostradamus trap.
Always under construction
A continuous series of decisions, informed by evidence, economics, risk, and changing circumstances. An internally built application may depend on a commercial model. That model may be replaced by a better one. A vendor may introduce functionality that makes part of an internal application redundant. A use case that did not make economic sense six months ago may suddenly be worth pursuing. Meanwhile, the business keeps changing.
Own, control what is yours. Build it if you must.
The thing to own is the business logic, the data, the workflow, and the judgment around it: the places where the value comes from knowledge, judgment, or your differentiated methods. A manufacturer’s quality rules and production exceptions. A logistics company’s view of capacity, customers, constraints. A services company’s way of turning expertise into repeatable work.
Whether or not you choose to build the AI and automation yourself is a separate decision, decoupled from ownership. You can have someone else build it while you retain ownership.
Buy to avoid reinventing the obvious
Buy what has become common capability or soon to become commodity. General drafting, search, document processing, coding assistance, and routine analysis don’t need recreation.
Use them where they help. Replace them when something better comes along. The software still has to fit the workflow, work with your systems, and behave sensibly with your data. A subscription does not absolve you of those responsibilities.
Wait when the cost of being wrong is higher than the value of being early
Wait when an incorrect answer could create a customer, safety, financial, or legal problem that your business is not prepared to handle. Wait when your data is unreliable. Wait when nobody clearly owns the process. Wait when the economics do not work. And certainly wait when the business case begins with, “We should use AI for something.” The decision to wait should be deliberate, not procrastination disguised as the default option.
Reversibility
This matters enormously when navigating such choppy waters. Move quickly where decisions are reversible. Move carefully where they are not. That requires more than selecting tools.
The mid-market’s particular predicament
You cannot afford to build everything. You cannot afford to buy indiscriminately, like larger enterprises might. And you certainly cannot afford to wait indefinitely. You should be able to say “Not yet” with confidence, that phrase seems to have temporarily gone missing in the AI lexicon.
You need a practical way to make these decisions close to the business, with enough technical judgment to avoid being led around by a relentless, continuous stream of confusing AI breakthroughs.
The business needs people who understand enough about the work to identify where AI can create measurable value, enough about the technology to distinguish durable capability from temporary excitement, and enough about the economics to know when to build, buy, pause, replace, or stop. They need to see whether the components work together. Whether the risks are controlled. And they need to call it when yesterday’s good decision is no longer effective.
Less prophecy. More judgment.
Technology is not going to slow down. If anything, it will accelerate. Our decisions, however, still happen at human speed. They have to.
“What will AI look like in five years?” is not a useful question for the business operator. The useful question is, “What enduring actions can I take now to manage through uncertainty I cannot wish away?”
Good judgment, made close to the business and revised when the facts change, is not a veneer over the chaos. It is the stability layer.
Build selectively. Buy selectively. Wait selectively.
Nostradamus, unfortunately, cannot make the next board meeting.
Comments