Series
Software that knows
Most software keeps what an organisation knows in code, where only developers can read it. This series tests a different idea: write the knowledge down as a model, and derive the software from it. Each part tests the idea on a real source, such as clinical guidance, a law or a popular diagram, and says how we would know it is wrong.
Start with part 1. The companion posts go deeper into one part; you can read them or skip them.
- Demos: lab.almilo.com
- Code: github.com/almilo/software-that-knows
Software that knows: the hypothesis
Business applications keep business knowledge implicit in code. A hypothesis: make it explicit and put it at the centre of the software.
Software that knows: why now
Formalising business knowledge used to be slow and expensive. LLMs make the first draft cheap. Review and accountability stay with people.
Software that knows: what knowing means
Data, information, knowledge, understanding and wisdom: what it means for software to know, from raw bits to judgement.
Software that knows: where knowledge sits
One fever rule, found in code, a database, configuration, an SOP and a clinician's head. And the alternative: one formal, explicit file.
Software that knows: from guideline to application
A working application built from published WHO guidance: where the knowledge comes from, the two models behind it, and what an agent does.
- Companion Software that knows, in detail: what's in a field
One input field in an ontology, from bare concept to translated question with a rule: what an application keeps in ten places, stated once.
- Companion Software that knows, in detail: from statements to a working form
How a generator of about 370 lines turns the statements of a knowledge model into a form, validation, rules and words, without knowing what fever is.
- Companion Software that knows, in detail: what's in a field
Software that knows: the knowledge model as grounding
A free-text note instead of a form: four interpreters, from plain words to LLMs, all grounded in the same knowledge model.
Software that knows: from a diagram to a decision
A popular comparison diagram, written down as facts, questions and its author's rules: an app that asks only what matters, decides and says why.
Software that knows: the rule nobody owns
A refund policy with six sources and no owner. Written down as a model, it cannot decide either, but it shows exactly what has to be decided.
Software that knows: one app, two diagrams
One generic app, a second diagram: the new domain is four files, no new code, and checks find two places where the drawing is ambiguous.
Software that knows: one model, three applications on three stacks
One knowledge model runs in three applications on different technologies. The knowledge stays when the application changes, and tests show it.
Software that knows: where the magic happens
A reference, a short prompt and a goal it can check: a coding agent gets almost there, and a final review takes it the rest of the way.