AI for building software (3)

P.S. I am sharing a rearticulation of point 6 from my first AI for building software post. As I reread it, I realized I was not clear.

Original: So who will be these external vendors who manufacture software? Many are throwing their hat in the ring: Cloud provider incumbents like Microsoft (GitHub Copilot), Google (Antigravity); Foundation model leaders (Anthropic with Claude Code, OpenAI with Codex); and newer dedicated vendors like Cursor. My sense is there will be enough space for everyone but differentiation will be hard to establish and being low cost will matter. Basic functionality and quality are, of course, assumed. You can get a standard white mug from $1–$1,000, and while one may be better than the other, they both work just fine as mugs.

Take 2

External vendors will set up factories to manufacture software. These factories will consist of:

1. Chips, power, water

2. Foundation models (external and in-house)

3. Software to manufacture software (i.e., agents & associated tooling)

4. Actual capital

5. Political capital to get these factories set up wherever it is most conducive to operate and to create positive PR.

Software factories will create both standard software and specialized software. Standard software (as simple as a login screen to as complex as some parts of a Salesforce CRM) is roughly equivalent to nail manufacturing today. Nails have been made in factories for decades. They range in price, design, and quality but a nail is a nail and the manufacturing process doesn't change drastically. And therefore most of this will be outsourced to software factories.

Specialized software will continue to be more in-house, or may be forced or lobbied to be in-house by regulators. Imagine the software that launches a nuclear missile array, does robotic surgery, or touches "secret sauce" data or systems like credit card spending. This will all be more in-house than outsourced, not because the AI agents can't write it—of course they can—but because companies won't risk a leak to their software factory.

There will be factories that position themselves as producers of standardized and / or specialized software. For standardized software, cost and delivery time will matter most, like it does with mugs today. For specialized software, quality and customization will matter, less so cost. As an additional aside: standardized software today is largely implemented by IT services vendors. AI is coming for them.

So who are the factory owners? 1. Cloud providers (Microsoft with GitHub Copilot, Google with Gemini/Antigravity): Exceptionally well-positioned. They have all of the inputs + distribution + money-printing machines in their houses.

1. Model providers (OpenAI with Codex, Anthropic with Claude Code): Well positioned. They have many of the inputs I wrote about.

2. Niche players (Cursor). They have the least number of inputs but will be the fastest and will have nothing to lose as they are startups.

Differentiation will be driven by either being a low-cost manufacturer (looking at you, Google, with your own models and TPUs) or being the highest-quality manufacturer, i.e., being the Germany for machinery/Switzerland for watches of specialized software manufacturing.