When Code Is Cheap, What Is Still Worth Building?

I have been experimenting with building small apps with AI coding tools, in the mornings and in the evenings. Things that would have needed a developer and a budget two years ago now take me a few weekends. That is exciting, and slightly unsettling, because everyone else can do the same.

So when I watched a recent video by Chris, a product designer who runs the YouTube channel Build Great Products, it answered a question I had been circling for months: if code is no longer scarce, what is?

01.Three Shifts

Chris argues that generic software, the dashboards and CRM tools of the last decade, will struggle as a business. He gives three reasons.

• Building got cheap. A generic product can be cloned in a weekend. The old protection, that software took years and large teams to build, is gone. In his words, the moat "is no longer the code".
• Agents became users. Software now has two customers: the person clicking through a screen, and the AI agent calling it through an API, a command line tool or an MCP server, with no screen at all.
• Pricing follows the work. Customers increasingly want to pay for an outcome, not for a number of logins. Software starts to look more like a service.

02.Eight Things Worth Building

From those shifts he derives eight types of product with real opportunity. In short:

1. Software-enabled services. Do not sell accounting software. Be the accounting firm that is ten times faster, with a small team and AI doing the routine work. He cites AI-native firms such as Crosby and Pilot.
2. Vertical agent harnesses. The model is general. The value is in the environment around it: the rules, tools, knowledge, workflows and memory for one industry. Harvey in legal is his example.
3. Headless AI software. Tools with no interface, built only for agents to call. Narrow, technical, usually priced by usage.
4. Agent-native systems. A plugin that runs inside Claude or ChatGPT, paired with a dashboard to track progress. Suited to coaches and consultants with a repeatable method.
5. Taste-led niche products. Built for a small community that trusts the maker, not for the mass market.
6. Ambient consumer agents. Assistants that live where people already are, such as group chats, rather than in a new app.
7. Proprietary data products. Anyone can copy an app. Nobody can copy data that sits outside what AI models were trained on.
8. Open-source core products. The code is free and works as marketing. The money is in hosting, support and paid tiers.

03.What This Means for Professional Services

I work in professional services, so the first two types are the ones I keep thinking about.

The software-enabled service is a direct challenge to how firms like mine are built. If a two-person team with AI can deliver what used to take twenty, the pyramid model changes. The people at the base of it, and how they learn the job, change with it.

But the framework also shows where established firms are strong. A vertical harness is, in practice, a firm's method written down: its checklists, its judgement calls, its view of what good looks like. Proprietary data is years of engagement experience. Distribution is client trust. Those are exactly the assets that took decades to build and that a weekend clone cannot copy.

The question for any expert, in a firm or on their own, is no longer whether to build software. It is which part of their expertise becomes the harness, and who owns it.

04.Where I Would Push Back

The video is a useful map, but a few points deserve caution.

• "A weekend clone" is true for generic tools, not for regulated work. Security, compliance, integration with client systems and liability still take time and money. In tax, legal or finance, that is a moat in itself.
• Outcome pricing is harder than it sounds. It works when the outcome is measurable. Much professional work is about judgement and risk avoided, which is difficult to price per result.
• The platform risk is real. Chris acknowledges this himself for headless tools: you may build something that a model provider gives away for free next quarter. The same applies to plugins built on someone else's marketplace.
• The source has a stake. The channel promotes its own community and product framework. That does not make the analysis wrong, but it is worth knowing.

---

What stays with me is the simplest line of the argument. When building is cheap, the scarce thing is knowing what is worth building and caring enough to build it well. For people with deep expertise in a narrow field, that is good news. The tools have finally caught up with what they know.