What You’ll Find This Week

The conventional wisdom is that regulated industries are where innovation goes to die. Heavy compliance overhead, multi-year approval timelines, agencies that move at government speed. Fast-moving companies build the future. Regulated ones wait for permission.

That framing has taken a hit in the AI era, and not for the reason anyone expected. This newsletter has covered the efficiency trap across a lot of territory: why innovation labs default to failure, why financial metrics quietly kill long-term R&D, why the CAIO absorbed an innovation mandate that used to live somewhere else. The trap keeps returning in different forms. AI turned out to be its most effective disguise yet.

The industries everyone dismissed as innovation graveyards ended up being the ones the shortcut couldn't reach. This week's article is about what that produced.

Here’s what you’ll find:

Don’t Miss Our Latest Podcast

This Week’s Article

Ask anyone in corporate strategy to name the most innovative companies in the economy and you'll get the same answer. The AI platforms. The cloud providers. The software companies that redrew industries in a decade.

Nobody names pharma. Nobody names banking. Nobody names medical devices.

They should.

The industries that everyone assumes are innovation graveyards, weighed down by approval timelines, compliance overhead, and regulatory frameworks that move in years rather than months, are producing the broadest portfolios of real invention in the American economy right now. Meanwhile, nearly every other industry is deploying AI and calling it innovation. You can track the evidence on any job board. "AI Innovation Manager." "VP of AI Innovation." "Chief AI Innovation Officer." None of these titles existed three years ago. All of them signal the same confusion between deployment and invention.

Those are not the same activity. And the gap between them is widening.

I wrote in the first issue of this newsletter that corporate innovation's golden age ended when financial metrics made efficiency more legible than invention. ROIC and RONA don't reward decade-long research programs with uncertain outcomes. They reward cost reduction. Companies didn't abandon invention because it stopped working. They abandoned it because the metrics stopped recognizing it.

That shift was well underway before AI arrived. What AI did was complete it.

When AI scaled in 2023, it offered something companies had been looking for: a credible way to simulate the discovery process without actually running it. Horizon scanning? AI. Competitive intelligence? AI. Rapid ideation, concept testing, early-stage prototyping, customer interview synthesis? AI. The exploratory work that used to justify a corporate innovation function, including what innovation labs were supposedly doing before they became builder decommissioning chambers, could now be compressed by an order of magnitude.

So companies did what the financial logic had been pushing them toward for decades. They stopped funding the exploration and outsourced it to a model.

This is where the story might end, if AI were just another cost-cutting tool. But AI arrived as something that looks like technology, and technology has always carried the implication of innovation. So this didn't feel like making the familiar efficiency trade. It felt like investing in the future. In May's issue, I documented what that feeling produced: the CAIO absorbed the innovation mandate, and acquiescence was nearly universal.

But AI deployment is efficiency work. It can't discover a mechanism of action that has never been tried. It can't tell you whether a new financial product will survive regulatory scrutiny before you build it. It can't invent a medical device category.

Unregulated industries had the option to confuse the simulation for the substance. They took it, practically in unison. The companies that spent a decade disrupting efficiency-obsessed enterprises made the same trade those enterprises made, and they did it willingly, because this time the efficiency play arrived wearing technology's clothes.

Regulated industries didn't have that option. The agencies they answer to don't accept shortcuts.

In financial services, speed without guardrails has a well-documented downside: the 2008 crisis. Basel III capital requirements, Dodd-Frank oversight, and CFPB supervision created constraints that couldn't be automated around. You can deploy AI to optimize a process that already exists. You can't prompt-engineer your way around a capital adequacy requirement.

Europe's PSD2 directive required banks to open their APIs to third parties. No incumbent wanted to do it. The regulation gave them no choice. Banks that invested in genuine API infrastructure built something that outlasted the compliance requirement. What followed was a direct product of the mandate: Monzo, Revolut, N26, the account aggregation and payment infrastructure that now underlies embedded finance across Europe. None of it would exist if regulators had left the incumbents alone.

In pharma, the FDA's approval process doesn't care how you found the molecule. It demands proof that it works, in humans, at scale, repeatedly. AI can accelerate candidate identification, and it has, dramatically. But it can't run the Phase III trial. It can't generate the safety data. It can't replace the clinical evidence the agency requires at every stage of the pathway. The constraint forces real work that AI can assist but can't replace.

Moderna's mRNA platform, developed under the full weight of FDA oversight, didn't just produce COVID vaccines. It's now in clinical trials for HIV, cancer, and cardiovascular disease. Not a faster version of something that already existed. Something new.

In healthcare, the FDA's quality bar is not a prompt. Digital therapeutics are applications that function as medical interventions and require regulatory clearance. They have had to produce actual evidence of clinical efficacy, in patients, before shipping. You can't submit a 510(k) clearance application for an AI wrapper on an existing protocol. You have to prove the thing works. That standard is what separates a product from a feature.

Christensen's original Innovator's Dilemma was about incumbents failing to pursue disruptive innovation because their existing business model was too profitable to abandon. The logic that made them successful became the logic that made them vulnerable. They couldn't see the threat coming, because it arrived as something small and cheap and not worth worrying about.

What's happening now has the same structure. The victims are different.

The efficiency logic this newsletter has tracked since issue one has found its most powerful instrument in AI: the same dynamic that traded long-term invention for short-term returns, that hollowed out corporate R&D and turned Bell Labs-style research into a budget line. It has also found its most ironic victims. Not the lumbering incumbents who couldn't move fast enough. The companies that built their entire identities on being different from those incumbents.

They made the same trade. They did it eagerly and fast. Because AI doesn't look like an efficiency play. It looks like technology. And technology, everyone assumed, is where innovation lives.

Efficiency killed corporate innovation the first time by making financial returns more legible than invention. Now it's doing it again, dressed in the most exciting technology of the decade. The companies that won the last era of disruption are walking straight into the same pattern that defeated the companies they disrupted. And they're celebrating the walk.

The innovation debt that accumulates here runs deeper than the residue of failed initiatives. It's the atrophy of a capability that was never exercised under pressure. When AI adoption becomes table stakes and everyone has the same efficiency gains, the competitive delta returns to product invention. The organizations that let the muscle go will discover they can't rebuild it on demand.

The pharma company that has been running clinical trials for five years under FDA oversight knows how to produce invention under constraint. The tech company that has been running AI pilots for five years under no external constraint has a different kind of knowledge, and a capability gap it won't notice until it matters.

There are two ways to read this, and both of them are uncomfortable.

If you work in an unregulated industry, the shortcut has a price that isn't showing up on any balance sheet yet. Right now, the AI deployment story is working. Efficiency gains are real. The cost savings are measurable. The narrative holds.

What isn't visible yet is what you're not building. The exploration that used to happen, even badly, even through innovation labs that mostly failed, was maintaining the organizational capacity to invent: the people who knew how to work toward an outcome they couldn't guarantee, the tolerance for uncertainty that makes discovery possible, the institutional muscle memory for pushing against resistance without knowing if you'd break through. AI deployment, by design, removes all of that.

When the efficiency gains compress across your industry and someone else finds a product insight that AI couldn't generate, you will want that muscle back. It won't be there.

If you work in a regulated industry, stop treating your regulatory environment as purely a drag. It is one: operationally, financially, administratively. The compliance burden is real. The timelines are punishing. None of that is wrong.

But the same constraint that costs you speed is preserving something your competitors are quietly losing: the organizational capability to produce invention under pressure. That capability is, right now, a structural competitive advantage in any market where the shortcut doesn't reach.

The risk for regulated industries is the same as for everyone else, just delayed. As AI tooling improves and regulators develop frameworks for AI-generated evidence, that shortcut will eventually arrive in your industry too. The FDA's evolving AI guidance, which has published multiple major frameworks since 2021 and issued new draft guidance as recently as January 2025, is already moving in this direction. The organizations that use the current window to build real innovation capability, not just compliance infrastructure, will be better positioned when it does.

The conventional wisdom that regulation kills innovation was always a partial truth. Regulation kills certain kinds of innovation: the fast, cheap, incremental kind that moves by removing friction and seeing what sticks. It doesn't kill the harder kind. It can't, because the harder kind is exactly what the constraint demands.

What nobody anticipated is that the arrival of AI would reveal this so clearly, so fast. The industries with the most freedom to move fast moved immediately to the shortcut. The industries with the least freedom kept doing the hard thing, because the hard thing was the only path to the outcome.

If you want to know where real invention is happening in 2026, don't look for the companies with the most AI press releases. Look for the ones whose regulators still require proof.

The guardrails aren't doing their job in spite of innovation.

They're doing it instead.

We've opened submissions for Discover an Innovator: a curated directory of innovation professionals who are available for what's next, whether that's a full-time role, a fractional engagement, an advisory relationship, or something they haven't been asked to do yet. The profiles section will launch once we have enough profiles submitted to make it worthwhile. (Which means I need your help!)

If you're a company looking for innovation talent with unusual range, this is where you can expect to find it.

If you're an innovation professional who's had trouble articulating your story on the platforms you're supposed to use, we'd like to profile you.

Submissions are open now 👇