← INSIGHTS

MPS BRIEF · SEPTEMBER 2026

A bad price doesn’t make it a bad technology.

AI capability, AI-company valuation and an operating company’s ability to capture value are three different questions. Treating them as one is where bad decisions begin.

By David Morris

4 MIN READ

I believe deeply in AI.

I also don’t believe every dollar chasing it is going to earn a return.

On Monday, AI-linked stocks sold off after Anthropic’s Dario Amodei — and then Sam Altman and Elon Musk — said the industry needs to slow the pace of frontier development.

That wasn’t the market’s only problem. Oil was up. Bond yields were up. Rate expectations were changing.

Still, the reaction felt telling. We’ve spent the last few years treating three very different questions as if they were the same.

THREE DIFFERENT QUESTIONS

What can AI do?

A capability question. Answered by using the tools, not by reading about them.

What are AI companies worth?

A pricing question. It moves on sentiment, rates and supply of capital — not on what the models can do for you.

How much of that value can the rest of us actually capture?

An operating question. It is decided inside your workflows, your data and your decisions.

I don’t have a clean answer to all three. I don’t think anyone does.

But I’ve used these tools enough now to know this: even if the models didn’t improve much from here, there is already more than enough capability to change how a lot of companies work.

What I’ve watched happen — both ways

Where it worked.

AI shortened research, prepared decision briefs, flagged pipeline risk and took hours out of recurring executive work. Nothing exotic — the same tasks every week, done to a defined standard, with someone reviewing the output.

Where it didn’t.

Companies bought the tools and changed almost nothing else. No outcome named, no workflow redesigned, no owner. The license renews. The operating model is identical.

That’s the gap I keep coming back to.

The technology really can be extraordinary. That doesn’t mean every investment assumption around it makes sense.

Where we start instead

At MPS, we start with the outcome someone wants to improve, then work backward. Five questions, in this order:

  1. What outcome are we trying to move?Named before any tool is chosen. If it can’t be named, nothing downstream will be measurable.
  2. What has to change in the workflow?AI added on top of an unchanged process produces the same result with more steps.
  3. Who owns it?A person, not a committee — accountable for the output being used, not just produced.
  4. Which data can be trusted?It doesn’t have to be perfect, but the definitions underneath it have to be understood.
  5. Where do people need to stay in the loop?Decided deliberately, before the first confident wrong answer, not after it.

And then the one that settles the argument: how will we know it actually worked?

That part is less exciting than a new model release, but it’s where the value actually shows up.

The question I’d put to you

If AI didn’t get materially better from here, what could your company still change with what already exists?

Market price, model capability and operating adoption are three different questions. Treating them as one is where bad decisions begin. If you have a view — including the view that the whole category is overbuilt — tell us where you land.

Selective by design.

SOURCES

  1. Reuters, September 14 — AI-linked stocks fell after leading AI executives called for slowing frontier development, while rising oil prices and rate expectations added pressure.

    reuters.com/business/ai-warnings-knock-nasdaq-futures-pressure-tech-stocks-2026-09-14
  2. A study covering more than 1,500 organizations and 17 million ChatGPT Enterprise messages found wide differences in adoption and concluded that firms are still actively learning how to integrate AI into organizational workflows.

    arxiv.org/abs/2608.12236

AUTHOR

David Morris

Start with the outcome, not the model.

Twenty questions, about three minutes, and a 0–100 readiness score with the three things to fix first.

Take the AI Readiness Assessment → See the C-Suite Agents →