How we use AI

How closely a human worked on each page and module

Work on fxEngine began in 2008, as the formula engine of Calcapp. Until the summer of 2025, we wrote and maintained the engine by hand. That includes most of the compiler, the runtime and the modules that generate our documentation, which date from 2021. We began using coding agents in the summer of 2025.

Today, we use AI tools, including LLM-based coding agents, throughout fxEngine. They help us write and test code, find defects and draft the documentation and the posts on this site. Every page on this site, except the landing page, names how much a human was involved, as one of three standards. The READMEs of our repositories apply the same standards to code modules.

Our formula reference is generated from the documentation in our source code, and each of its pages is labeled by how its text was written. A label describes how a page or a module was written. Every page and every module is written under our direction.

In short

  • Delegated. A human decides what the work is for and checks the results, but doesn’t read all of the work itself. We expect pages written to this standard to be very rare.
  • Reviewed. A human reads every change and sends it back until it is correct.
  • Authored. The work is the human’s own. This standard differs between pages and code. A human writes the draft of a page. Code may be written by an agent, but the human has read and revised every line.

Delegated

A human owns what the work is for and whether it delivers. An agent owns the work itself, which a human is not expected to read in full.

For code, a human writes the requirements and checks the results, while the agent writes the source code and any tests. This suits temporary or internal tools whose source code nobody else sees. For a page, a human checks parts of the text, such as a long list generated from data, but not every sentence. We expect pages written to this standard to be very rare.

Reviewed

A human steers the work and reads every change, sending it back until it is correct and fits what surrounds it. The result may still read as AI-assisted work that a human has reviewed, rather than as the human’s own writing.

For code, the human knows the whole codebase and steers its architecture. For a page, the human checks every claim and every sentence before it is published.

Authored

The work is the human author’s own, and the author stands behind every line of code and every sentence. What that takes differs between pages and code.

For a page, the author writes the draft. An agent may then help revise it, suggest a better word, catch a spelling mistake or even draft a passage, but the writing is principally the author’s.

For code, an agent may write all of it, documentation included. It is accepted only if the author has read and revised every line, until the code reads almost exactly as it would have if the author had written it by hand.

Keeping the labels true

A label describes a page as it is now. If a later change to a page falls short of its standard, the label changes with it. If parts of a page were written to different standards, the page names the lowest of them.

Questions

Write to us at support@fxengine.io.

Reviewed by a human

fxEngine Shell Want this in your own terminal? A standalone version is coming.
Loading fxEngine Shell

fxEngine Shell for your terminal

A version you install on your own computer is coming. Subscribe to the newsletter to hear when it's ready.