People take the decisions. AI writes the code.

A traditional build of an enterprise system is long for two reasons, and neither is the difficulty of the business. The first is that every project writes its foundations again: logins, permissions, audit, approvals, reporting. The second is that the code for the business itself is typed by a large team, and every misunderstanding found late is paid for in rework and change requests.

Both have changed. On Phlo the foundations are already written and already running. And AI now writes the code for the part that is yours. The band of a project that used to take months, the build, becomes one of the shorter ones: 15% of a first build.

What did not shrink is the part that was always the real work. Three fifths of a first build is spent before any code is written: on site, on the process map and on the specification. That work is done by people, in person, and no model does it for them. Building enterprise systems in weeks, not quarters, comes from removing the typing, not the thinking.

Writing code is the part of a software project that AI now does well. Understanding an operation is not. Nobody can prompt their way to knowing that a farmer arrives at one in the morning with no contract, or that the packing floor weighs its offcuts and converts them back to metres.

So the work is divided along that line. People visit the site, map the process, design the solution and take every decision. AI writes code. The result is a team that delivers at least 3x faster than top tier IT services firms, without handing the thinking to a model.

A general-purpose model asked to build a business system will guess at a structure, and guess differently each time. Ours does not start from a blank page. It is trained on the framework itself, so it builds the way our engineers build.

The framework leaves little room to improvise:

• One way to record a change. Every action in a Phlo system is an event written through a single path. There is no second route for code to take.
• One pattern for a process. A new process is a set of events, the rules that decide when each may be recorded, and the screens that record them.
• Tests that walk the whole journey. Automated tests follow a record from its first step to its last and check real values. A change that breaks one does not ship.

Every screen, action and data field is agreed with you before a line of code is written. That matters more with AI, not less. Code is now quick to produce, so the cost of a project is no longer in typing it. It is in building the wrong thing.

A precise specification is also what AI builds from. That is what removes the change-request cycle that stretches a traditional project.

The specification outlasts the build. Together with the process map, it is the written model of your operation: its entities, its events and its exceptions. Every later system is built from it, whoever builds it.

The time saved does not come off the engagement as a discount on attention. It goes on site, with the people who run the process, learning where the exceptions are. On your floor shows what that time finds.