I've spent a career building software, and I have never seen the cost of making it fall the way it has in the past two years.

A feature that once took a team a quarter now takes one developer and an AI assistant a week. A product that took years to build can be copied, convincingly, in months. Sometimes it's simply given away. That changes the question every software vendor should be asking. It's no longer "how good is our product?" It's "why would anyone stay with us when they can get the same thing somewhere else, cheaper or free?"

My answer: because you answered the phone.

The code is no longer the moat

For decades, the hard part of software was building it. That difficulty protected vendors. If you'd spent ten years and hundreds of engineers on a product, a competitor had to do the same to catch you.

AI has drained that moat. Even the AI models themselves aren't safe. In February 2026, Anthropic said three AI labs had run "distillation" campaigns against its Claude models, using around 24,000 fraudulent accounts and some 16 million exchanges to train their own models on Claude's answers. Those are Anthropic's allegations, but the point stands either way: if a frontier AI model can be copied by asking it enough questions, your invoicing app, your photo editor or your CRM certainly can.

Once the product can be matched, it stops being the reason customers stay.

Adobe had the luxury of lock-in

Large software companies have enjoyed a comfortable position for a long time. Customers needed the product, there were few real alternatives, and leaving was painful. So the business model drifted from serving customers to holding on to them.

Adobe is the textbook case. In June 2024 the US Federal Trade Commission took action against Adobe and two of its executives, alleging the company buried an early termination fee of 50 per cent of the remaining payments on its popular "annual, paid monthly" plan, and made customers navigate numerous pages to cancel. In March 2026 Adobe agreed to a settlement worth US$150 million, half as a payment and half as free services for affected customers. Adobe disagreed with the government's claims and denied wrongdoing.

Whatever the legal outcome, every Creative Cloud customer who has tried to leave knows how it felt. And now look at what's happened to the moat that made that behaviour possible.

In October 2025, Canva made Affinity completely free: photo editing, vector illustration and page layout in one professional app, a direct alternative to Photoshop, Illustrator and InDesign. Add the free tools that already existed and AI tools that generate in seconds what used to take an afternoon in Photoshop, and the "you have no choice" era is over.

In March 2026 Adobe announced its long-serving CEO would step down, and its shares fell to a three-year low amid worries about AI disruption and competition from the likes of Canva and Figma. I can't imagine the conversations in that boardroom. The product is still excellent. But excellent no longer means irreplaceable.

Why do billion-dollar companies hide from their customers?

This has always mystified me. Companies making billions have made reaching a human something to be discouraged. You email a "no-reply" address. You press 1, then 4, then 9. You argue with a chatbot that can't do the one thing you need. Eventually you give up, which, I suspect, is the point.

I understand the reason. Public companies answer to investors, and investors want return on investment every quarter. Support staff are a cost line, and cost lines get cut. A phone tree is cheaper than a person.

But customers notice, and they say so. A Gartner survey of nearly 5,800 customers found 64 per cent would prefer companies didn't use AI for customer service, and 53 per cent would consider switching to a competitor over it. Their biggest fear wasn't the technology. It was that AI would make it even harder to reach a person.

of customers would prefer companies didn't use AI for service (Gartner)

64%

would consider switching to a competitor over it (Gartner)

53%

Adobe settlement over subscription fees and cancellation (March 2026)

US$150m

Even the companies most excited about AI have learnt this the hard way. Klarna replaced much of its service team with AI, then in 2025 its CEO said it was hiring human agents again so customers could always talk to a person, after a focus on cost had hurt quality.

Service is the product now

Here's my bet for the next few years. When software is cheap to build, customers will choose vendors the way they choose a good mechanic or accountant: on trust, responsiveness and whether someone actually listens.

That means:

• A human answers the phone. Not after six menu options. Not "your call is important to us." A person who knows the product and can fix things.
• Use AI to help your staff, not to hide them. Let AI draft replies, find answers and handle the routine work, so your people have time for the conversations that matter.
• Make leaving easy. If customers stay only because cancelling is painful, you don't have customers, you have hostages. They'll leave the moment a free alternative appears, and they'll tell everyone why.
• Listen and act. Feedback that disappears into a ticket queue is worse than no feedback form at all. Close the loop and tell people what changed.
• Be honest about price. No buried fees, no surprise renewals.

None of this is new. It's what good small businesses have always done. The difference is that it's now the main thing separating you from a competitor who can rebuild your product over a long weekend.

Welcome to the one-person billion-dollar company

There's a bigger shift underneath this one. Big public companies exist partly because building big things used to take big capital and lots of people. AI is eroding both requirements.

The number of US listed companies has fallen about 25 per cent since 2000, according to SIFMA, and the US averages far fewer IPOs a year than it did in the 1990s. Sam Altman has talked about tech CEOs running a betting pool on the first one-person billion-dollar company.

We may already be close. In April 2026 the New York Times profiled Medvi, a telehealth business run by founder Matthew Gallagher and his brother, which reportedly made US$401 million in sales in its first full year and projected US$1.8 billion for 2026, using AI for code, marketing and customer service. Those are the company's own figures and a forecast, not audited results, and critics have questioned its marketing practices. Tellingly, its AI support bot reportedly invented drug prices and products that didn't exist. That's the warning: AI can scale a business, but it can't care about the customer for you.

I believe we'll see fewer publicly listed companies and many more small, owner-run ones that never take a cent from outside investors. Without a board demanding quarterly returns, those owners can make a different choice: spend the money on people who look after customers.

That's the opportunity. The giants built their walls with product and lock-in. AI has knocked the walls down. The businesses that win next won't be the ones with the best code. They'll be the ones where someone picks up the phone.