Academy Guide

AI Fundamentals for Platform Builders

A practical introduction to language models, prompts, context, reliability, cost and where AI fits inside a real product.

AI should support a workflow

The strongest products use AI to improve a defined task. The weakest simply place a chat box over a model and hope the user discovers value.

Academy principleStart with the user outcome. Decide where AI adds speed, personalisation or judgement support only after the workflow is clear.

Choosing a model

Select models against the actual task: reasoning quality, latency, structured output, context capacity, multimodal support and cost. A more expensive model is not automatically the correct production model.

  • Define the task and required output.
  • Create representative test cases.
  • Compare quality, latency and cost.
  • Test failure modes and ambiguous inputs.

A useful prompt structure

ROLE
You are a domain-specific assistant.

CONTEXT
Use the supplied user profile and product rules.

TASK
Produce a concise recommendation.

CONSTRAINTS
Do not invent missing facts. Ask one focused question when required.

OUTPUT
Return valid structured JSON.

Reliability is a product problem

Prompts alone do not create dependable systems. Good platforms add validation, retrieval, deterministic rules, structured outputs, guardrails, observability and clear fallbacks.

Control cost deliberately

Store reusable results where appropriate, reduce unnecessary context, use smaller models for simple tasks and measure cost per successful user outcome—not merely cost per token.

Next steps

Continue with the platform architecture guide, then download the AI Prompt Specification from the resource library.

Continue to architecture →