When API Isn’t an API: Why AI Tools Need Your Context

AI Fluency
Health Research
Moving from computer science to malaria research, I keep reading familiar acronyms the wrong way. AI tools make the same mistake, and the fix is the same: give them context.
Author

Kaiza Ilomo

Published

November 5, 2026

I trained as a computer scientist. Today I work in malaria research at Ifakara Health Institute, and I expected to learn new science there. I didn’t expect to relearn my own vocabulary.

Same letters, different meaning

When a malaria colleague mentions a district’s API, my brain still reaches for Application Programming Interface. They mean Annual Parasite Incidence: confirmed malaria cases in a year per 1,000 people. When they say MIS, I think Management Information System; they mean the Malaria Indicator Survey.

It doesn’t stop there:

Term In computer science In malaria research
API Application Programming Interface Annual Parasite Incidence: confirmed cases per 1,000 people per year
MIS Management Information System Malaria Indicator Survey: a national household survey on bed nets, treatment and malaria testing
PR Pull request Parasite rate: the percentage of people in an age group with detectable parasites on a given day
CI Continuous integration Confidence interval
Vector An array of numbers, as in a “vector database” The mosquito that carries the parasite

Even inside malaria research, the letters slip. Some papers expand API as Annual Parasite Index, and it is easy to mix it up with the parasite rate, a different measure: one counts new cases over a year, the other is a snapshot of who is infected on one day (WHO malaria terminology).

My current work brings together climatology, entomology, epidemiology, hydrology and demography datasets, and every one of those disciplines has its own shorthand.

AI tools have the same problem

An AI model reads your words the way I read them on day one: by guessing the most common meaning. The text these models learn from mentions software APIs far more often than parasite incidence, so with no other clues, “API” will usually be read as software.

The obvious mix-ups are easy to catch. An answer about endpoints and JSON when you asked about malaria cases is clearly wrong. The dangerous mix-ups are the plausible ones. Ask a tool to “compare API across districts”, and if it treats API as a parasite rate, you get a fluent, confident answer that compares the wrong thing: a percentage on one day instead of cases per 1,000 people over a year.

The model isn’t being careless. It is working with what you gave it.

Give the context a new colleague would need

In What Is AI Fluency? I described Description: give an AI tool the context, constraints, examples and format you would give a new colleague. Acronyms are where this matters most. A few habits help:

  1. Say what field you’re in. One line, such as “I work in malaria epidemiology in Tanzania”, makes the right reading of every acronym that follows far more likely.
  2. Spell out acronyms on first use. It’s the rule journals already ask of us: “API (annual parasite incidence: confirmed cases per 1,000 people per year)”. Add the definition when a term has variants.
  3. Keep a glossary the tool always sees. Most AI tools let you save standing instructions (custom instructions, project instructions, or a file such as AGENTS.md for coding agents). Put your field’s acronyms there once instead of retyping them in every chat. This website’s own repository has an AGENTS.md file for the coding agents that help maintain it.
  4. Check which meaning came back. Ask the tool to list the definitions it assumed before it answers. That’s Discernment: judge an answer by its meaning, not by how fluent it sounds.

The same tools that misread acronyms can also help you learn them, if you ask in context: “In malaria epidemiology, what does API mean, and how is it calculated?” Check the answer against an authoritative source, such as WHO’s terminology, before you rely on it.

Not only an AI problem

Ambiguity is a communication problem first. The habits that help an AI tool help people too: define your terms in reports, code comments and meetings with colleagues from other fields. AI tools just show the cost of missing context faster.

Learn more

← All posts

Back to top