From prompts to agents in three months

AI Fluency is one programme in three months: 24 live sessions, two a week, and three finished projects. Each month has six teaching sessions and then one project built across two days. You can take one month, two or all three, and each month builds on the one before.

Month 1: foundations, and your first agent

Who it suits. People who already use AI, paste in a prompt, get something usable and move on, with a quiet sense that others are getting far more out of it. Month 1 is where everyone begins.

What you learn.

1. AI Introduction. What is genuinely new about today's AI, where it is strong and where it is unreliable. You bring the problem you most want to hand to an AI system, and you return to it all programme.
2. Basics of LLM. Enough of how a large language model works to predict its behaviour: tokens, context windows, why the same prompt gives different answers, and where hallucination comes from.
3. AI Prompt Engineering. Prompting as engineering, built on three ideas: context (what the model can see), loop (iterating towards a good answer) and graph (breaking work into connected steps). You learn to diagnose a poor result by asking which of the three went wrong.
4. AI Hygiene. Working responsibly: what never goes into a prompt, checking output before it reaches a decision or a customer, attribution and disclosure, and habits that hold up when you are rushing.
5. Automation, AI agents and agentic work. Three things people often confuse, and which problems each one suits.
6. Action Planning. Turning a problem into steps a system can carry out, deciding what stays with a person, and defining a good result before you build.

What you build. Sessions 7 and 8 are the First Line Agent: one project across two days. Day one gets your agent running end to end, however rough. Day two fixes what broke, handles the cases that fail, and ends with a demonstration and review. You leave with a working agent you built from a standing start.

1. 1.Day one: get it running

   End to end, however rough.

2. 2.Day two: fix what broke

   Handle the cases that fail.

3. 3.Demonstrate and review

   Show what you built and get feedback.

Month 2: how agents actually work

Who it suits. People who have a first agent running and want to know why agents fall over in their second week, and how to build ones that hold up.

What you learn.

1. Anatomy of Agents. The parts every agent has, how a request travels through them, and where failures usually sit.
2. LLM Preprocessing. What happens to your documents before a model sees them: getting them into usable form, how you split them, and why that changes the answers.
3. Vector Databases. How machines store meaning rather than words, and why the closest match is not always the right one. You set one up over your own content.
4. RAG Fundamentals. Grounding a model in your own material, why retrieval systems give confident wrong answers, and how to trace the cause.
5. API Wrappers, Tools, MCP. Where an agent stops talking and starts acting, connected to real systems.
6. Agent Memory. What an agent should remember between turns and between runs, and how to keep that memory useful as it grows.

What you build. The Agentic Workflow, again across two days. You bring retrieval, tools and memory together into one system, then make it hold up on inputs it has not seen. You leave with an agent that remembers, retrieves and uses real tools.

Month 3: systems that run without you

Who it suits. People who want the whole arc, including the part most never reach: getting an autonomous system evaluated, kept inside safe limits and deployed.

What you learn.

1. Multi-step Reasoning. Working through a harder problem in steps, keeping the run on track, and knowing when the extra cost and time are worth it.
2. Multi-Agent Systems. When several agents beat one and when they do not, how work passes between them, and how to keep the whole thing observable.
3. Agentic Frameworks. What frameworks give you, what they hide, and how to build so the framework can be replaced.
4. Agentic Evaluations. How you know whether an agent is any good: deciding what to measure, building a test set from real cases, and running it every time the system changes.
5. Safety and Guardrails. Where an autonomous system can do real harm, how to limit what it may do, and what must always stay with a person.
6. Custom Toolkits. Building and documenting tools of your own for the final project.

What you build. The Autonomous Loop, deployed. Day one builds a loop that runs unattended, with the guardrails and evaluation from this month. Day two, "Optimise and Deploy", tunes it for cost, speed and reliability, deploys it so it keeps running after the session ends, and sets up monitoring. You finish with three working systems across the programme.

1. 1.Month 1: first agent

   Foundations, then a First Line Agent.

2. 2.Month 2: agentic workflow

   Retrieval, tools and memory in one system.

3. 3.Month 3: autonomous loop

   Evaluated, guarded and deployed.

How the tiers map to the months

Innovator is where the work becomes certifiable: it qualifies you for the AI Fluency exam. Architect includes exam access and the Upfyn AI Internship on the live Upfyn project, offered when you complete it; Builder and Innovator learners can reach it through the free monthly internship exam and an interview with the Board. The internship comes with finishing the work, never with a purchase alone, and it does not guarantee a job.

Every session is live, in a monthly group of 25, and every project is reviewed by a senior engineer. Compare the tiers and see the full syllabus on the AI Fluency page, or read how project sessions work.