Every technology gets cheaper the longer you wait. That is the most reasonable argument for staying on the fence about AI, and it is half right. The tools will keep getting cheaper. What will not get cheaper, at least not for a few more years, is the experience of using them well.
That gap, between tools anyone can buy and experience very few companies have, makes this the riskiest and the most promising stretch of the AI revolution so far.
That gap, between tools anyone can buy and experience very few companies have, makes this the riskiest and the most promising stretch of the AI revolution so far. It is also why this is the moment when outside expertise is worth the most.
The Half-Written Playbook
Expertise is worth the least at the two ends of a technological revolution. At the start, nobody has done the work yet, so advice is mostly speculation. Near the end, the playbook has been written so many times that it ships with the software. In the middle, the tools are powerful, the playbook is half written, and the people who have done the work in many companies carry knowledge that is not written down anywhere yet.
That is where AI stands today. The middle is also where the stakes run highest, because the tools have moved from interesting to consequential. Describing OpenAI’s latest business usage data to Semafor this summer, the company’s chief economist, Ronnie Chatterji, said businesses are shifting toward “work that’s more doing rather than asking.” A tool that drafts an email can embarrass you. A tool that approves an invoice, answers a customer, or changes a record can cost you.
Peak Opportunity, Peak Risk
The payoff is concentrating quickly. PwC’s 2026 AI Performance Study, a survey of more than 1,200 senior executives across 25 sectors, found roughly a fifth of companies capturing about three-quarters of AI’s economic value. PwC warns that the gap is likely to widen, because the leaders keep learning faster. The OpenAI data shows how fast they are moving. In January, the most active tenth of OpenAI’s business customers used about 2.6 times as much AI output per user as the median company. By June, it was 8.3 times. Chatterji compared the moment to the early internet, when everyone else caught up once the gains were proven. They caught up to where the leaders used to be.
The risks are rising just as fast. When MIT’s NANDA initiative studied enterprise AI for The GenAI Divide: State of AI in Business 2025, it found 95 percent of organizations getting no measurable return from generative AI. The researchers traced the gap not to the technology but to how companies went about using it, and approach is exactly what experience improves.
Some of the doors are closing, too. The procurement and IT leaders MIT interviewed in the first half of 2025 expected enterprises to lock in AI vendor relationships that would be “nearly impossible to unwind” within about 18 months. Most of that time has now passed. Systems that learn from a company’s data and workflows build switching costs every month they run. That is good news if you chose well, and an expensive lesson if you did not.
One-Way Doors
Jeff Bezos once sorted decisions into one-way doors, which are hard to reverse, and two-way doors, which are not. Most of the AI decisions with closing windows are one-way doors: which platform learns from your data, which vendors you commit to, which workflows you rebuild first. Those are exactly the decisions where experience pays for itself. The decision to get an experienced second opinion before you walk through them is a two-way door.
Staying on the fence gets this backwards. It treats the reversible decision as the risky one, while the irreversible ones get made anyway, by vendors, by employees, and by competitors.
What Outside Expertise Buys
Outside expertise is a way to borrow experience instead of earning it one mistake at a time, and it buys three things that are hard to build on your own.
When the window is short and the stakes are high, the scarcest resource is not money or software. It is time to learn. Outside expertise is a way to borrow experience instead of earning it one mistake at a time, and it buys three things that are hard to build on your own.
The first is pattern recognition. An advisor who has watched the same decisions play out in other companies knows which uses pay back, which pilots stall, and why. Your team would learn those lessons the expensive way, on your data, with your customers watching. That fits what the MIT researchers found: AI initiatives built with specialized outside partners reached full deployment about twice as often as those built entirely in-house.
The second is independence. Every vendor in your stack now has an AI story. One CIO interviewed for the MIT study summed up the flood: “We’ve seen dozens of demos this year. Maybe one or two are genuinely useful.” Telling the one or two from the rest is hard from inside a sales process, and much easier for someone who sells no software at all.
The third is calibration. Most leadership teams cannot tell from the inside whether they are ahead or behind. You see your own company in detail and your competitors mostly through their press releases. Someone who works across many companies can tell you where you actually stand.
Tools in a Master’s Hands
AI is a tool, and tools amplify whoever holds them. A power saw makes a master carpenter faster and an amateur more dangerous. The economists Ajay Agrawal, Joshua Gans, and Avi Goldfarb made the deeper point in Prediction Machines: as machines make prediction cheap, human judgment becomes more valuable, not less. The tools are becoming a commodity. The judgment about where to point them is not.
None of this means every use of AI needs an expert in the room. Your people should be experimenting with everyday tasks, and the faster they learn, the better. But important work, the kind that touches customers, regulators, or the way you make money, belongs in practiced hands. A master does not slow that work down. A master makes the first steps smaller, safer, and cheaper to learn from.
If you do bring in help, choose it the way you would choose a surgeon, by how many times they have done this exact work before. Look for someone who has held the job rather than studied it, who knows an industry like yours, and who sells no software, so the advice carries no second agenda.
A Challenge for the Fence
If you are reading this, something probably put outside help on your mind. Maybe it was a board member’s question you could not fully answer, or a competitor’s announcement that sounded like a head start. Maybe it was a vendor demo that seemed too good to be true, or the discovery that your team was already using AI tools nobody had approved. Whatever it was, that was your judgment speaking, and it deserves more weight than the voice telling you to wait another quarter.
So here is a challenge. If you are on the fence, do yourself a favor and listen to the gut feeling that made you consider tech consulting in the first place. Start small, with an assessment or one month of advisory service. Both are two-way doors. If you don’t see the value in that month, at least you will know something worth knowing: you are already at the leading edge.
Either way, you will be off the fence.
The tools are becoming a commodity. The judgment about where to point them is not.
The tools are becoming a commodity. The judgment about where to point them is not. If you want an experienced second opinion before your next AI decision becomes a one-way door, consider starting small, with a tech/AI assessment or one month of advisory service.
