I had an idea this week for a business I’m not going to start. Take one of the hosting platforms people already deploy to, and have it sell the hard parts: phone numbers, SMS, video calls, email that lands in the inbox. Click a button and get a working service on the same bill as your database, so someone vibe coding a product can build all of it in one place, including the bits that touch the real world.

That idea appeals for the same reason most SaaS founders should be nervous. If you can describe a feature, someone can now build it in an afternoon. The things that are still hard to build are almost never features. They’re the infrastructure underneath, the licences around it and the people inside it.

“A bit nicer” stopped being a business

For fifteen years the standard pitch for a new SaaS product was some version of “we do what the incumbent does, but nicer”. Cleaner interface, faster onboarding, one less click. I’ve designed a lot of those products and I’ll defend the work. A nicer product did win customers, because building it took a funded team a year and the incumbent couldn’t copy it fast enough.

That year is gone. A capable person with Claude Code or Cursor can rebuild the surface of most CRUD apps in days, and copy your best interaction an hour after seeing it. Being a bit nicer only ever lasted as long as it was expensive to copy, and it isn’t expensive any more.

I run a design studio, so I’m an odd person to be saying this. Design still decides whether people stay once they’ve arrived. What it can’t do now is stop a competitor shipping the same screens next month.

A technical moat has to be deep or it doesn’t count

The usual reply is “we have proprietary tech”. Look closely and it’s often a clever prompt chain, a data pipeline or a well-tuned search index: things a good engineer built in a month. If a model and a motivated team can reproduce it in a quarter, you have a head start, and head starts get spent.

Deep means work that can’t be compressed by throwing more agents at it. Years of operating history. Hardware. Knowledge you only get by breaking things at scale and fixing them. Plenty of companies have cloned Stripe’s API. None of them got Stripe’s bank relationships, fraud data and card-network connections along with it.

The real world is still hard

Try sending a text message from a brand new product. You need a number, a provider that won’t get your traffic flagged as spam, opt-out handling, and in a growing list of countries some form of sender registration. In the US a business has to go through 10DLC registration before it can text customers at volume. Very little of that is code. It’s paperwork, relationships and reputation, and an agent can’t do any of it for you.

Video is the same. Getting two people on a call is a WebRTC demo. Keeping that call up on hotel wifi across three continents means relay servers in the right regions, codecs negotiated per device, and someone awake at 3am when a region degrades.

That’s why the idea has stuck with me. The best hosting platforms won developers by making deploys boring. The next layer is making the physical stuff boring: numbers, messages, calls, deliverability. Whoever turns those into a button on the platform people already build on owns the part of the stack vibe coding can’t reach. And because it’s billed by usage, it grows with every customer that succeeds on top of it.

Regulation is a moat you can’t prompt your way through

We see this at Hill, where I lead product. Hill is software, and we work hard on it. But what customers are paying for is a product that’s allowed to exist. It operates inside a regulated perimeter, and getting inside that perimeter takes licences, compliance work and time. A competitor could copy every screen we have. They can’t copy our permission to run them.

Regulated industries used to look like the worst places to start a software company: slow sales, heavy overheads, lawyers in every meeting. Those same overheads now make them some of the best.

Networks still compound

The other moat that survives is the oldest one. If your product’s value is the other people on it, copying the product gets you an empty room. LinkedIn has been out-designed many times and the graph hasn’t moved. Marketplaces, messaging, anything where a user’s contacts, history or reputation lives inside the product all get stronger as cheap clones multiply, because the clones split the newcomers between them and the incumbent keeps everyone who’s already there.

What I’d build now

If I were starting a SaaS company this year, I’d ask one question of the idea first: which part of this couldn’t a good engineer with an agent rebuild in a month? If the honest answer is the design or the feature set, I’d keep looking.

The answers that hold up all have weight to them. A licence, a carrier relationship, hardware in a rack, data nobody else is allowed to hold, a network of people who are already there. Get one of those first. Then make the software on top as nice as you can, because people still notice, and because it’s the part you can now build fastest.