The Business of Software

I don’t know the future, but I believe most people aren’t thinking this AI thing through.

Let’s talk about the business of software.

The free market argument for housing affordability is to remove red tape and make it easy to build. There’s no need to force developers to include “affordable” units in new projects. The idea is simple: if we build new luxury homes, older luxury homes become cheaper, and middle-class homes cheaper still. As a thought experiment, if a solar flare broke all modern cars, would crappy old beaters become cheaper or more expensive?

It’s all about supply and demand.

So, if simple software, like personal apps and internal tools, becomes free, what happens to slightly more sophisticated software? Presumably it becomes cheaper. But where do we draw the line? Five years ago you could build approximately nothing vibecoding. Today, virtually anyone can build an internal tool or personal app. A dedicated vibecoder with enough practice can build a simple SaaS application.

Have we reached the steady state? Where will we be in another five years? Will we have more software or less? Will software be easier to build or harder? Will we have more or fewer people that can build software? What does that mean for the business of selling software?

Furthermore, how many use cases that formerly relied on software can now be solved by an agent doing throwaway work? Do I need a personal financial planning tool, or do I just throw some data into Claude and ask for help?

You tell the agent what you want. The agent does it until it gets stuck. You don't know or care how it's doing it, so long as you can trust the results. The agent keeps getting better every 6 months, for free. You don't need to write more software or better prompts. Just keep doing your job and the thing will improve.

— Agents are a Generalized Technology

I’m not saying I understand the implications for today’s large SaaS companies. Software is one component of building a successful business. But today we have organizations whose main value prop is building and selling software. The software is the end. Maybe that will shift to firms more strictly selling solutions to problems, done through software.

Think of a business that sells me a tool to do performance reviews. Are they directly solving a problem? The problem is that I want to promote the best and fire the worst. Do I need software to administer the managerial process of performing reviews? Or do I need an AI that tells me who to promote and who to let go? Do you see the difference? Does Salesforce bring me sales, or is it just software to administer the human process of sales?

The businesses that survive will be the ones building the tools agents need to actually solve the problem:

If someone creates a critical tool and integrates it with a frontier LLM to achieve valuable tasks, customers will willingly pay a premium for tokens resold by the tool-builder. But these tools aren't trivial MCP servers executing granular tasks. Instead, they involve large data, complex requirements, high compute needs, distributed systems, and advanced simulations.

— It's the Tools

Software is not dead. It’s about to become orders of magnitude larger. What does that mean for the business of software? Does it stay the same? Unlikely.