Part 1 of 4. Your business’s velocity of code generation is way less important than how quickly it can evaluate if that code generates value.

Every few decades, businesses develop a new way to make learning more disciplined, more repeatable, and more useful. The Toyota Production System brought continuous learning into manufacturing. Agile shortened the loop between building software and understanding customer needs. DevOps shortened the loop between deploying software and learning from it in production.

AI Software Factories are being pitched as the next entry on that list. But the way I’ve heard AI Software Factories described doesn’t seem a big enough revolution to earn that place yet, and the reasons it falls short matter more than the technology itself.

The trap of measuring production

The current conversation about AI often begins with production metrics: how many lines of code a model can generate, how many tickets an agent can complete, how many deployments a team can ship, how many tokens an engineer can burn through in a day. These numbers matter operationally, but they measure activity, not value.

I have watched companies generate enormous amounts of code and still build the wrong product. I have watched teams deploy continuously and never come closer to understanding their customers. I have watched businesses automate every step of delivery and accelerate themselves away from the market in the process.

How quickly a business can produce a feature matters less than how quickly it can discover whether that feature is valuable.

Every business is testing a belief

Every business exists to create value for customers. But no business begins with perfect knowledge of what customers need, what they will pay for, or how the market will respond.

Every strategy is therefore a set of beliefs about reality.

Every product is an expression of those beliefs.

Every release is an intervention in the world.

Every customer interaction is a source of evidence.

Every business result is feedback about the quality of the business’s understanding.

The basic pattern is simple:

We believe that doing X for these customers will produce Y.

The business then acts, observes what happens, and updates its belief. If the product creates value, the business learns what to continue. If it does not, the business learns what to change, or discovers that its question was poorly formed in the first place.

Products are not merely outputs. They are instruments for asking reality a question.

Value is the reduction of meaningful uncertainty

Asking reality questions and learning from the responses lets a business build a model of its customers’ values and the shape of the market. The more accurate that model is, the better decisions the business makes and the better it becomes at achieving the outcomes of its actions. Value comes from reducing that uncertainty, not from action by itself.

The learning engine can be represented as:

Action → Evidence → Knowledge → Better Decisions → More Valuable Action

Action without evidence is motion. Evidence without interpretation is data. Knowledge that never changes a decision is inert.

The economic value appears when learning improves what the business does next: when it builds a better product, serves a customer more effectively, avoids waste, reduces risk, or discovers a more valuable opportunity. The scarce capability is the velocity and integrity of validated learning, not raw production speed.

Every era in this lineage had to meet that standard. Not every one of them met it fully. Some needed additions or clarifications later. But a revolution in how businesses work can’t just mean working faster. It has to mean working better.