Some Companies Are Already on Life Support. AI Just Hasn’t Pulled the Plug Yet.

I have started thinking about a certain class of established software company as being on life support.

That sounds harsher than I intend it to be.

These companies may have plenty of revenue. They may be profitable. Customers may genuinely love them. Renewal rates may look terrific. In healthcare, financial services, government, and other regulated markets they may hold certifications, integrations, procurement contracts, decades of data, and what look like impenetrable barriers to entry.

They can still be on life support.

The mistake is assuming a company is alive because customers are still paying it.

If you started this company today, knowing what is possible with AI, would you build anything resembling the company you currently operate?

If the answer is no, you have a problem. And adding an AI button to the old product is not going to solve it.

Intercom Is the Case Founders Should Study

Intercom is interesting because the sequence is almost too neat.

Co-founder Eoghan McCabe stepped away as CEO in 2020 and returned in 2022. The board did not bring him back because of generative AI; ChatGPT had not launched yet. Intercom was working through slowing growth and strategic problems of its own.

Then ChatGPT arrived, and leadership figured out quickly that this was not another feature cycle.

The company has described that moment starkly. Its own retrospective says management wondered whether AI was an existential threat or the company’s best chance of survival. Intercom formed a cross-functional effort around AI, cancelled non-AI projects, and committed roughly $100 million to replatforming the business around it.

That distinction matters.

• They did not create an AI innovation committee.
• They did not park a chatbot next to the existing product and call themselves AI-enabled.
• They changed what the company was.

Intercom later described the transformation this way: over roughly three years it rebuilt almost everything it had spent the previous eleven years building. By May 2026 the transformation had gone far enough that the company renamed itself Fin. Intercom survived as the name of the customer service platform; Fin became the company and the declared future. McCabe also said they had finished rebuilding the underlying help desk as “Intercom 2.”

A successful SaaS company looked at eleven years of work and decided that eleven years of work was not necessarily an asset. Some of it had become baggage.

Founders May Have to Destroy What Founders Built

Workday gives us another signal. In February 2026 co-founder Aneel Bhusri returned as CEO, and the company explicitly framed his return around its next chapter in the AI era. Bhusri called AI a bigger transformation than SaaS. Internally the effort was later described as a “re-founding.”

I like that word. It is much better than transformation.

Corporate transformation usually means keeping the company intact and changing enough around the edges to survive the next planning cycle. Re-founding asks a far more dangerous question: what if the assumptions we founded this company on are no longer true?

Workday was once the disruptor. It attacked an earlier generation of enterprise software using the architectural and economic advantages of SaaS. Now its founder says AI is a larger transition than the one his company rode in on. That should make every mature SaaS founder uncomfortable.

If AI is bigger than SaaS, converting a SaaS product into AI-powered SaaS is probably not enough. Imagine responding to Salesforce in 1999 by putting a web page in front of a client-server CRM product. Technically, it would have been on the web. Strategically, it would have missed the entire point.

A lot of companies are doing the AI equivalent of that right now.

AI-First May Prolong the Disease

We need better language here, because three very different postures are currently sharing one vocabulary.

• AI-augmented: AI makes your existing people more productive.
• AI-first: AI becomes the default assumption when designing products and operations.
• AI-native: the company, product, architecture, workflows, and operating model are designed around what AI can now do rather than around what software and labor used to cost.

The first two are useful. Neither one gets you to the third.

Shopify gives a clean operating example. Founder and CEO Tobi Lütke told teams that AI use was now a baseline expectation, and that managers asking for additional headcount should first demonstrate why the result could not be achieved with AI. That inverts the default. Most companies still ask how many people we need to do this. The new question is why we need another person to do this.

Electric went further. Founder Ryan Denehy relaunched the company around an AI platform for IT and HR automation. Electric began as an IT help desk business; the rebuilt company uses what it learned across roughly two million historical support interactions as raw material for an AI-driven operating model. Denehy has described the process plainly as rebuilding Electric as an AI-native company.

That is the pattern I am interested in. Not AI features. Corporate self-cannibalization.

This Is the Incumbent’s Version of Native Alpha

I write elsewhere about Native Alpha, which is the advantage already sitting inside a company that would be hard or expensive for someone else to reproduce. This whole argument is the uncomfortable, incumbent-facing side of that idea.

Here is the part that stings. A great deal of what mature software companies count as advantage was never native to them. It was borrowed from an era when software was expensive to build and labor was expensive to deploy. Long implementation cycles, configuration systems, training departments, professional services, integration teams: those were advantages because they were hard, and they were hard because capability was scarce.

Capability is cheap now. Advantage is not. When the cost of capability collapses, every advantage that existed only because capability was expensive collapses with it, on a delay, quietly, while the renewals keep landing.

Some of what you call a moat is just the price somebody else used to have to pay.

The advantages that survive are different in kind: the relationships, the regulatory knowledge, the data nobody else has, the position in a workflow, the trust that took fifteen years to earn. Re-founding is how an incumbent goes looking for the advantage that survives cheap capability, instead of assuming the old one still holds because the invoices still clear.

Regulation Will Slow This Down. That Is Not the Same as Stopping It.

Healthcare founders should pay particular attention, because healthcare software has enjoyed an extraordinary amount of structural protection.

• Regulation protects incumbents.
• Integration complexity protects incumbents.
• HIPAA protects incumbents.
• FDA requirements protect incumbents.
• Hospital purchasing departments protect incumbents.
• Epic protects incumbents.
• Long implementation cycles protect incumbents.
• Fear protects incumbents.

For years this let mediocre software stay economically healthy far longer than mediocre software survives in consumer technology. It also produces a comfortable and possibly wrong conclusion: no startup can replace us because healthcare is too complicated.

Maybe. But consider the other possibility. Healthcare being complicated may be exactly why an AI-native company eventually holds such a large advantage.

Legacy software exists because humans could not manually track all the rules, workflows, documents, codes, permissions, exceptions, and handoffs involved in care delivery. So we built screens. Then more screens. Then configuration systems. Then workflow builders. Then implementation consultants. Then training departments. Then integration teams. Then customer-success teams explaining the software to the people who were supposed to use it.

AI does not eliminate regulation. It may eliminate a surprising amount of the machinery we built to help humans cope with regulation.

That is a very different proposition than a prettier interface, and it is worth several quarters of your attention.

The Competitor May Look Like Your Customer

Even established healthcare technology companies are noticing the shift. Mike Novotny founded clinical-trial software company Medrio in 2005, led it for about fifteen years, stepped away, then returned to run it again in 2025, and Medrio now publicly lists him as founder and CEO. Timing alone does not prove Medrio is tearing itself down around AI, so I would not stretch the comparison with Intercom. It is another signal that founders of mature regulated-technology businesses are coming back at an unusual technological moment.

The more aggressive examples come from companies with no legacy product to defend. Radley describes itself as an AI-native radiology practice. What caught my attention is that the founders are not only selling AI software to radiologists. They are acquiring a radiology practice and say they intend to rebuild the clinical and operational stack with agents from the ground up, arguing that point solutions cannot fix an operation built around obsolete systems and processes.

That is the threat incumbents should sit with. The competitor may not look like another software vendor. It may look like your customer. An AI-native radiology practice can eventually compete with a radiology software company, a services company, a staffing company, and another radiology practice at the same time. The category boundaries stop holding, which is considerably more dangerous than somebody shipping a nicer application.

Your Installed Base Is an Asset, or an Anesthetic

This is where successful founders get trapped.

You have ten thousand customers. They generate data. They give you distribution. They provide domain knowledge. They hand you real-world edge cases a two-person startup will not see for years. That is an extraordinary position, and Intercom used exactly that: it kept its customer relationships, domain knowledge, and support data while rebuilding around a fundamentally different technological model. Fin now says its products serve more than thirty thousand companies and handle millions of conversations a week.

That is how incumbency is supposed to work. But the installed base can also work as an anesthetic.

• Every customer has another requested feature.
• Every salesperson has another deal that needs a customization.
• Every large account has another integration you must not break.
• Every product manager has another roadmap.
• Every engineering manager has another migration that would be too risky.

Each argument is individually rational. Collectively they can kill the company.

Nobody decides to preserve the legacy company. Everybody merely makes sensible decisions that preserve it.

The Board May Be Part of the Problem

This is one reason founders matter here. Professional management is usually hired to operate the machine: grow revenue, protect margins, reduce churn, make the quarterly number, keep employees, keep customers, avoid unnecessary disruption.

Those are excellent instincts. They are terrible instincts when the machine itself has become obsolete.

Imagine telling a hired CEO: I want you to spend $100 million rebuilding something customers already pay us hundreds of millions of dollars to use. It sounds irresponsible, and under normal governance it is.

A founder can sometimes say something different. I built this. I know why we built it this way. Those assumptions no longer hold. We are going to build the thing that kills it.

That does not make founders better CEOs. Plenty of founders should never return. It means technological discontinuities occasionally require authority that ordinary corporate governance is specifically designed to suppress.

Do Not Kill the Old Product Tomorrow

This is where the argument gets misread. I am not suggesting you walk in Monday morning and delete the repository. Especially not in healthcare, where the software is load-bearing for people’s care.

The clever strategy is not destruction. It is controlled competition.

1. Create a small team.
2. Give them no obligation to preserve the existing architecture.
3. Do not require feature parity.
4. Do not ask them to migrate the current product.
5. Give them the same customer problem the company was founded to solve.
6. Ask them what they would build if they were founding this company this month.
7. Put it in front of real customers and measure it.
8. Let it compete with your existing product, and do not protect the incumbent.

Treat this as an R&D experiment long before it becomes a migration program. If the experiment fails, excellent, you learned something for the price of a small team. If it starts producing an embarrassingly simpler product with one developer, a handful of agents, and a fraction of your operating cost, pay very close attention.

That embarrassment may be the most valuable product research your company has done in years.

The Uncomfortable Question

Run this exercise honestly. Imagine your most dangerous competitor has no existing code, no current org chart, no sales compensation plan, no professional-services revenue to protect, no implementation consultants, no legacy pricing model, no roadmap, and no architectural review board explaining why the old architecture must survive.

They do have your domain knowledge. They understand your customers. They have modern foundation models, AI coding harnesses, and agent infrastructure. And they start today.

What would they build?

Now look at your own product. If the two look reasonably similar, you are probably fine. If they look radically different, I would not spend much time celebrating your renewal rate.

Some Companies Aren’t Dying. Their Assumptions Are.

There is a serious counterargument, and I hold it myself. AI has not destroyed enterprise software nearly as fast as the enthusiasts predicted. Large incumbents retain enormous advantages in data, integrations, trust, distribution, and customer inertia. Recent reporting on Salesforce, ServiceNow, Workday, and other enterprise vendors makes exactly that point.

I agree with it. It does not weaken the argument. It makes the trap more dangerous.

Disruption rarely arrives on the schedule predicted by the people most excited about it. The old product can remain useful for years. Mainframes are still running. COBOL is still running. Fax machines still exist in healthcare.

Regulation, switching costs, and recurring revenue can keep the patient alive for a surprisingly long time. They cannot make the patient young again.

That is why I use the phrase life support. The company can keep functioning long after the design assumptions underneath it have become invalid, and the functioning is what convinces everyone that nothing needs to change.

Founders Have One Enormous Advantage

The incumbent founder holds something the AI-native startup does not: customers, data, reputation, domain expertise, capital, distribution, regulatory knowledge, integration knowledge, and years of mistakes already paid for.

That is an extraordinary collection of assets, but only if you are willing to separate those assets from the product that accumulated around them. The opportunity is not to preserve the software. The opportunity is to preserve everything valuable the company learned while becoming willing to destroy the machinery it used to learn it.

Intercom seems to have understood this unusually early. Others are starting to.

So my advice to founders of mature healthcare and regulated-industry companies is simple enough. Do not wait until an AI-native company proves your architecture is obsolete. Prove it yourself. Put your current company in competition with the company you would start today, and then make sure the new company wins.

If you won’t kill your old product, eventually somebody who has never loved it will.