THE AI THAT EATS ITSELF

The word on the street is that AI is eating consulting’s lunch. I don’t buy it. Not the way it’s being sold.

AI is not eating anything. It has no mouth. It has no mind. It has no plan. What we call artificial intelligence is just very fast pattern matching. It is built on the work people have already done, the choices people have already made. It can copy. It cannot understand. A better name for it is assisted inference. Not intelligence.

And here is the part worth your time. The firms that have been selling AI the hardest are turning out to be its first victims. Look at what just happened to one of my global best 100 companies.

On June 18, Accenture (ticker ACN) hit the whole consulting world, and itself, hard. But you have to read the report carefully, because it tells two stories at once. The quarter that just ended was fine. It was the outlook that scared people.

Go down that table slowly. Earnings beat. Sales grew. Cash poured in. The company still paid out 2.2 billion to shareholders in three months. By the old way of keeping score, this was a strong, healthy business. And yet the stock had its worst day ever and is now down more than half this year. It trades at about eleven times earnings, a low it has not touched in years.

CEO Julie Sweet pointed to two reasons. One was the war in the Middle East, which she said cut about 100 million from sales and around 400 million from new orders as clients held back. The other was the bigger worry hanging over the whole trade: if AI tools can do the work in less time, clients need fewer paid hours.

Here is the irony. Accenture is the firm that is supposed to help clients put AI to work. Instead, the fear that AI kills demand for that very work has cut the stock in half. And watch this small detail: the company has stopped reporting its AI order number on its own line. That was the number investors were using to size up the opportunity. When a firm quietly puts away the scoreboard, the market writes in its own number. Usually a worse one.

This is not a one-off. A senior voice at Apollo said this month that consulting is next on AI’s chopping block, right after software. Rivals Capgemini and Infosys are each down more than 30% this year. The whole group is hurting. And don’t let Accenture’s 104 giant deals (100 million or more) fool you into calling a bottom. The trophy deals can hide a slow rot in the mid-size work underneath, and the mid-size work is exactly where cheap software replaces junior hours first.

Let me be plain here, because sloppy use of the word intelligence is costing people money.

Good consulting is thinking. A good adviser looks at a problem nobody has seen before, weighs who wants what, makes a judgment call when the answer isn’t clear, and then owns the result. That is reasoning. It is human.

Today’s AI is guessing the next step, very fast, based on everything that has been written and done before. Think of it as a map drawn from our own past work. No feelings. No goals of its own. No loyalty to you.

That is not an insult. Fast guessing at this scale is powerful. It is quick and it is cheap. Because it runs on the record of millions of past jobs, people can no longer out-type it on ordinary, repeat work. A machine that has read a million past projects will draft the standard report faster than a team of ten over three months. For the routine middle of the consulting business, that is the whole ballgame.

But the machine runs thin on anything new. When the problem has no match in what it was trained on, when people’s interests truly clash, when somebody has to be on the hook for a wrong call, copying does not help. There is nothing to copy. That is the ground people still hold. The error investors are making is to assume the machine has won the whole field, when really it has won the easy, predictable middle and left the hard edges alone.

My view, and it is the view of someone who puts protecting capital first, is that Accenture gets through this once it learns to use AI for what it really is.

The firm is not weak. It threw off 3.6 billion in cash in a single quarter. It has a strong balance sheet. It still wins the biggest, most tangled jobs on the planet, the kind that need real people to run them. In the same week it cut its outlook, it spent about 4.2 billion to buy control of Dragos, plus runZero and NetRise, all in cybersecurity, moving into protecting the systems that run power grids and factories. That is a firm climbing toward harder work and away from the easy middle, toward jobs where trust, judgment, and being on the hook still earn a fee.

That is the right move. The model that is dying is selling human hours by the thousand. The model that lives is using cheap AI as the shovel and charging for knowing where to dig. The winners are not the ones selling shovels. They are the ones digging faster than everyone else, and a lot of them won’t need a big consulting contract to tell them how.

So the caution runs both ways. I would not buy the drop in consulting as if this were just another dip in a normal cycle. And I would not assume AI consultants are safe from AI. A strong balance sheet is not the same as a strong moat when the moat is being filled in by software that costs a fraction of a junior analyst. But a profitable leader at eleven times earnings, with the cash to remake itself, is being priced as if it cannot change at all. I think that is too harsh. It is a reaction, not a verdict. The number I will watch is new orders. When they turn back up, the case is made.

This is a good moment to show how I would actually handle a name like this, because Accenture is a perfect lesson in why one good number is never enough.

Start with quality. On my first gate, a company has to score 80 or higher on the Ziggma fundamental rating, my screen for low-risk strength across growth, profit, value, and financial health. Accenture scores near the very top. So it sails through gate one. It is a strong, durable business. But look what just happened to the stock. That is the whole point. Quality tells you what to own. It does not tell you when.

That is the work of the next gates. I read my INSTAT score alongside RSI and MACD, and together they mark both ends of the move. Near a floor, when the selling looks spent, they point to an attractive price to buy. Near a ceiling, when the upside looks spent, they point to an attractive price to sell. A 99 quality score does not override that reading. A great company at the wrong price is still a bad trade.

Then the last gate, which no machine does for me. I read the Point and Figure chart myself before I risk a dollar. The tools narrow the field. I make the call. That is the line between guessing and judgment, and it is the same line this whole piece is about.

Everything above assumes today’s AI stays what it is: fast guessing. But that is not what its makers are promising. The pitch for the next round is human-level reasoning, machines that don’t just guess the next word but actually think a problem through. This is worth a cool, honest look, because if it is real, it changes the Accenture story and a lot more.

Two things are true at the same time right now. First, the new “reasoning” models are not a marketing trick. They are built to break a problem into steps, try a few different paths, and check their own work instead of blurting out an answer. On step-by-step tasks, that cuts mistakes. Researchers, including some inside the AI labs, have found real signs of this kind of structured work going on inside the models.

Second, and just as important, nobody has settled whether this is real understanding or just very good imitation. That argument is still wide open among serious people. Even the studies that find structure inside the models are careful to add that it works in ways quite different from how a person thinks. The machine acts like it reasons. Whether that counts as understanding depends on what you mean by the word, and that fight is far from over.

My rule here is the same one I use on any story that sounds too good. Keep what has been shown apart from what has been promised and never let a promise carry a position. What has been shown is that fast step-by-step AI now handles a wider band of ordinary brain work, which is exactly the squeeze on the middle of consulting. What has been promised is a jump to real reasoning that would squeeze the top end too. The first is happening now. The second is a forecast, and it is shouted loudest by the people selling it.

So I hold both ideas at once without flinching. The shift is real and it is speeding up, and any firm built on billing routine human hours is on the wrong side of it. At the same time, the consulting that survives is the judgment, the responsibility, and the fresh thinking that today’s machines still can’t do, and may never do. Until human-level reasoning is shown to me instead of announced to me, I will treat it as a label on a map, not the ground itself.

The real question for Accenture is not whether it bounces back. It is what business it is even in now. If the answer is selling human hours, the tide is against it. If the answer is owning the hard, high-trust work that is tough to automate, and using cheap AI to do it faster, then this drop will look, years from now, like the day the market mixed up a business in transition with a business in decline. I think it is the first. The tide is digital. The clock now runs in fractions of a second. But the firms that understand the gap between guessing and thinking, and price their work to match, are the ones still standing when the noise dies down.

I held a licensed fiduciary investment manager designation (formerly NASAA Series 65) and retired in June 2026. This piece is for information and education. It is not personal investment advice, and it is not a call to buy or sell any stock. The figures here come from public reporting on Accenture’s latest quarter. Do your own homework and think about your own situation before you act.