Learn it in three minutes.
One concept per piece. A plain-language explanation, the steps that matter, and the one thing people get wrong. If it can't be said in three minutes, it becomes a deep dive instead.
Tokens, context windows and why your bill exploded
What a token really is, why a 1M context window isn't free, and the two settings that halve your spend.
What actually happens inside a RAG pipeline
Chunk, embed, retrieve, re-rank, generate — and the three places it quietly goes wrong in production.
Agents vs. workflows: when you don't need an agent
A decision tree. Most "agentic" features are a state machine with better marketing.
Evals: how to know your model got worse
Golden sets, LLM-as-judge, and why "it feels better" is not a release gate.
Fine-tuning vs. RAG vs. prompting
Three tools, three different problems. Most teams reach for the wrong one first.
What "multimodal" actually changes for your product
Images, audio and documents as first-class inputs — and the new failure modes that come with them.
MCP and why tool-calling standardisation matters
One protocol between models and your systems, instead of a bespoke integration per vendor.
Hallucination is not a bug you can patch
Why fluent invention is intrinsic to the method, and what to design instead of a fix.
Prompt injection, explained without the hype
Why your agent will follow instructions hidden in a PDF, and the four controls that actually help.
Zero trust in one diagram
Never trust, always verify — what that actually means for an identity-first architecture.
Privacy-by-design for AI features
The seven questions to answer before a single user record touches a model.
The OWASP Top 10 for LLM apps, in plain English
The threat list every AI PM should be able to recite from memory.
Data residency and why your EU launch slipped
Where data physically sits, who can access it, and the questions to ask a model vendor first.
Threat modelling an agentic feature
Half an hour with a whiteboard that saves an incident review.
Output vs. outcome vs. OKR
A one-page test for which one your roadmap is actually measuring.
The user story that survives refinement
Acceptance criteria engineers don't rewrite, with a before-and-after on a real example.
North star metrics without the gaming
How to pick one number without letting the team quietly optimise the wrong thing.
How to say no to a stakeholder and keep them
Trade the request against something visible instead of defending a queue.
Writing a one-page PRD
Problem, evidence, non-goals, success measure, open questions. Anything longer is a document nobody finishes.