What AI Can't Do Today Won't Protect You From What it Will Do Tomorrow

What AI can't do today won't protect you from what it will do tomorrow.

There's an idea making the rounds right now: as long as AI is better at 90% of a job but not all of it, the remaining 10% makes the human (non-AI) contribution massively more valuable. That 10%, the argument goes, is your experience, your judgment, your trust, your taste, your creativity. In effect, it is your business's protective "moat."

It's a comforting idea. I understand why it's spreading. I even like a lot of it.

So let me be precise about what I'm not saying. I'm not saying that 10% doesn't matter. It does — it may be exactly the ground you need to defend this year. What I'm saying is that the 10% is a moving target, and treating a moving target like a permanent fortification is how good businesses get caught flat-footed. The moat isn't wrong because it's worthless. It's wrong because it won't hold still.

Here's the problem: most of what I'm reading right now treats AI as a story about function — about which specific tasks a machine can and can't yet do. And plenty of business owners are quietly taking shelter in that word "yet," clinging to whatever sliver of activity hasn't been absorbed, as if the sliver itself were the strategy.

Notice that word. "Yet." It's doing a lot of work.

AI will keep absorbing function. Not eventually, in some hand-wavy sense — but steadily, quarter over quarter, in nearly every service-delivery context you can name. That's not some bold prediction requiring courage; it's the trajectory we've already watched play out for two years running. Which means two things follow, and neither is comfortable:

One: whatever you've identified as your 10% — do not count on it lasting. Today's untouched sliver of judgment and taste is next year's prompt template. It may still be the fight you need to fight this year. Just don't confuse this year's fight with a permanent position.

Two: defending the sliver as a fixed position was the wrong strategy in the first place. Building your plan around "the stuff the tool can't replicate yet" means your competitive position is defined entirely by someone else's roadmap. You don't own that ground. You're just squatting on it until the tool gets there.

The Question Was Never About Function

Here's the thing that gets lost when we frame this as human-vs-machine: this is the same question you've always faced with any competitor, tool-based or not. Nobody builds a durable business by asking "what can my rivals not yet do?" That's a rearguard action, not a strategy. You build a durable business by asking the only two questions that have ever mattered: How do we make our service irreplaceable? How do we make buying from us the always-smart choice?

AI doesn't change those questions. It just raises the stakes of answering them honestly, because it strips away much of the padding — the friction, the paperwork, the "well, at least we're easier to work with than the alternative" — that let mediocre answers survive for decades.

Friction vs. Failure — And Why It Matters Here

I've written before about the difference between tools that solve friction and tools that solve failure.

A friction solution makes something easier. A failure solution makes something genuinely possible that wasn't possible before. In a market moving as fast as this one, only failure solutions hold durable value — friction gets commoditized almost immediately, because someone always builds a faster, cheaper way to remove the same friction.

The 90/10 framing is a friction argument wearing a failure costume. It says: AI removes friction from 90% of the job, and the human absorbs what's left. But "what's left" isn't automatically valuable just because a machine hasn't reached it yet. If that 10% is filigree — nice, but not something the customer would notice or mourn if it vanished — then the market will do exactly what markets do: it will re-price the whole job at the AI's rate, and quietly decide your 10% was never worth paying for. 

The 90% becomes the new 100%, and you're the one left explaining why your invoice looks the way it does.

If you're old enough, you remember the rapid transition from full-service travel agencies to online travel systems. Early versions were brutal — but the switch happened in 2–3 years. In that case, the "60%" at one third the cost became the new 100%.

This is the exact trap in resting on "AI can't exactly duplicate what I do." That sentence describes your process. It says nothing about your customer's priorities. And it is only ever your customer's priorities that set price and drive buying decisions.

This Isn't Theoretical — It's Already Happening

Four service industries are living this erosion in real time, and each one shows a different face of the same moving target.

Legal. For years, contract review and NDA analysis were the paralegal's and junior associate's bread and butter — the kind of judgment-heavy first pass that seemed safely human. Then AI platforms started beating the benchmark. In one widely cited study, AI reviewed a set of standard NDAs in seconds and caught more issues than a room of experienced attorneys working the same documents by hand. Major firms have used similar tools to cut M&A due diligence timelines by roughly a third.

The "10%" that document review lawyers assumed was theirs — pattern recognition, clause-by-clause diligence — turned out to be squarely in the 90% AI eats first. The lawyers who are thriving didn't try to hold that ground; they moved up into privilege calls, negotiation strategy, and client counsel. And before you say it — yes, some of that is just the next sliver on AI's roadmap. That's exactly the point. These lawyers didn't win by finding permanently safe ground. They won by proving they could move — and by holding the one thing no software release will ever ship: a human being who is personally accountable to the client for the call that gets made.

Accounting. Bookkeeping and routine compliance work was the reliable base of the small accounting firm for generations — necessary, billable, and assumed to require a trained professional. It's now commoditizing fast enough that industry benchmarking shows firm margins compressing even as revenue grows, because clients increasingly won't pay a premium for work they can see software doing.

The firms holding their margins aren't the ones arguing that a human still has to touch the ledger. They've repositioned around outcome-based advisory work — what the numbers mean, not who typed them in. Will AI eventually explain the numbers too? Count on it. The firms that survive that wave won't be the ones who bet that advisory was permanently safe. They'll be the ones whose clients would rather hear hard news from their accountant — a person who knows the business, the family, the history — and who kept moving up as the ground beneath them shifted.

Staffing and recruiting. Sourcing resumes and screening candidates against a job description is exactly the kind of high-volume, pattern-matching task AI was built for, and adoption data shows agencies using it are pulling meaningfully ahead of the ones that aren't.

But the more telling shift is on the buyer side: enterprise clients are actively consolidating their agency rosters and pushing out firms that show up with a list of resumes and a fee schedule, in favor of partners who can own the full talent lifecycle. The agencies protecting their 10% — "we still find better candidates" — are losing the relationship to agencies that redefined what they're irreplaceable for.

Construction and contracting. This one plays out differently than the first three — and it's worth pausing on, because construction's human core isn't a 10%-of-function story at all. It's physics. You can't automate a foundation pour or a framing crew out of existence, and nobody's pretending otherwise. That's a genuinely different kind of protection — but notice what it protects. It protects the crew. It doesn't protect the business.

Because the ground is still shifting underneath them. AI estimating and takeoff tools are now cutting bid-prep time by roughly 60%, letting the firms that use them chase 40–60% more opportunities without adding a single estimator — and industry data shows those firms winning a meaningfully higher share of the bids they submit. By some counts, well over half of the top 100 general contractors had adopted these tools by 2025.

The contractor working from field notes and a spreadsheet is now bidding against one who can price three times as many jobs, more precisely, in a fraction of the time — and losing jobs to them before the crew ever gets mentioned. The physical work AI can't touch never gets the chance to win the job. The 90% around it — the speed and precision of turning a plan set into a number — now decides whether a contractor is even in the room.

Same story, four industries: the ground everyone assumed was safely human turned out to be the fastest ground to move — or, in construction's case, the fastest ground to decide who even gets to compete. The firms still standing didn't win by proving AI couldn't quite match them. They won by moving when the ground moved. And yes — by embracing AI for everything that's become functional table stakes.

But Isn't That Just Another 10%?

If you've been reading carefully, you've spotted the objection: every one of those examples describes firms that moved to ground AI hadn't reached yet. But, you’re thinking: AI is still moving. So isn't "privilege calls" or "advisory work" just a different 10% — a fresh sliver waiting its turn on the roadmap?

Yes. It is. And if your takeaway from this article is "so I just need to find the right sliver," you've missed the point entirely.

Follow the logic all the way down, and you hit bedrock: on function alone, there is no permanently safe sliver. The only thing AI cannot ultimately touch is you.

That sounds philosophical. It's actually intensely commercial. Because there are two very different things that can defend a 10% differentiation, and most owners never separate them:

Capability — you can do something the machine can't. This is rented ground. It sits on someone else's product roadmap, and the lease expires without notice.

Preference —the customer could get it from the machine, and chooses you anyway. I can cook a very good salmon dinner at home. Sometimes I'd rather have someone make it for me — set the table, pour the wine, clear the plates. I'm not paying for capability; I'm paying for the experience of not doing it myself. Some banks figured this out years ago: as everything moved online, a few went deliberately retro — a greeter at the door, extended hours — and built a brand on the human touch precisely because it was no longer necessary. And nearly everything taught in a four-year degree is available in books and self-paced courses for a fraction of the tuition. Families write the checks anyway. It’s not about capability, it’s about belief. 

And here's the discipline this distinction imposes: stop romanticizing your gut. 

We all like to think our judgment is some unreplicable superpower. In reality, your gut is thousands of repetitions that have cut grooves into your thinking — pattern recognition running at high speed, for a human. Put enough computing power behind the same body of decisions, and the machine gets there. Judgment-as-function is absorbable. What isn't absorbable is what your presence does for the client: it gives them the confidence to actually act. Any good peer advisory roundtable proves this. Give AI two hours, and it will generate 47 strategic options as good as anything the room produces. What the room produces that the model doesn't: the conviction to choose one — the energy, the pushback, the read of the moment, the out-of-the-blue comment that turns out to be the whole answer.

Preference is a belief, and beliefs can shift — so it's not a set-and-forget moat either. But it's ground you own, because it attaches to you and your relationships, not to a task list. Capability rents. Preference, earned and re-earned, is the closest thing to owning.

The Moat Is a Practice, Not a Position

So should you map your 10% on a ten-year horizon instead — find the thing AI won't reach until 2036 and build there? Nobody can do that. Not you, not me — not even the largest technology companies on earth can genuinely plan a ten-year AI window for their own products. They can produce the document, put it in a binder (a virtual binder, naturally), and check the box that the ten-year planning was done. That's not knowledge. Nobody knows. Anyone selling you certainty about what AI will and won't do in 2031 is selling exactly that — opinion wrapped as certainty, not insight.

Which leads to the real conclusion: if the ground won't hold still, the durable asset isn't the ground. It's the muscle. The organizational habit of asking, every single year:

• What gap closed this year? What did we consider "safely ours" twelve months ago that a tool now does at 90% quality for 10% of the cost?
• What are we actually defending — capability or preference? If it's capability, what's our move when the lease expires?
• What would our customers say they'd miss? Not what we'd say. What they'd say.

Continuous reinvention used to be the ambitious option. Now it's the necessary commitment for continued survival. It's simply what staying in business requires.

The Right Question, Reframed in I³ Terms

So the real diagnostic isn't "what can't AI do." It's a different question entirely, and it's the same one that should sit underneath every value proposition you've ever built: what value — that my customer values, not what I personally think is impressive — am I delivering that can't be matched by an AI solution?

Years ago I developed a framework for exactly this diagnostic — I³, short for Innovative, Indispensable, Inspirational — and if anything, it's become more relevant now than when I first wrote it, because AI has removed the room for a business to coast on merely being competent.

Run that question through the I³ lens, and it gets sharper:

• Innovative isn't "I still do something AI hasn't automated." It's "I create something new enough that the comparison to AI doesn't even apply."
• Indispensable isn't "the client hasn't found a way around me yet." It's "removing me breaks something the client actually depends on to receive the value they were promised."
• Inspirational isn't a personality trait bolted onto competent execution. It's the reason a client keeps choosing you even after the friction you used to remove has been automated into irrelevance. It's where preference lives.

Notice what all three have in common: none of them are about function. They're about position — the same ground you'd need to hold against a scrappy competitor, a price-cutter, or a private-equity-backed roll-up. AI is just the newest, fastest-moving entrant into a game industrial owners have always had to play. Ok — it's the fastest-ever entrant into your competitive space. Same principles apply. Even more so.

What This Means Monday Morning

If you run a professional or business services, contracting, or trades business selling into larger enterprise customers, the question isn't "how much of my team's work can AI replace." That question has an answer, and the answer is: more than you'd like, sooner than you'd like, and it doesn't matter — because it's the wrong question.

The right question is the one you should have been asking all along, tool or no tool: What do we deliver that our customer would notice the absence of, in their business, in their outcomes — not in our process?

And here's the amendment AI forces on that old question: you don't get to answer it once. You answer it every year, because the ground under the answer keeps moving.

If you can't answer it in a sentence, you don't have a moat problem. You have a value proposition problem, and AI just made it visible faster than your last three competitors did.

The tool (or function) is never the strategy. It never was. AI hasn't changed that principle — it's just running out the clock on everyone who was hoping it might.

Here's a place to start, and it costs you nothing but honesty: write down, in one sentence, what your customer would notice missing if you disappeared. Then go ask three of them if they'd write the same sentence. The gap between those two answers is where the real work begins. 

And put a note in your calendar to do it again next year. The sentence will change. That's the point.