The full AI agent course curriculum

7 live workshops plus Demo Day across four weeks.
Stop dabbling with AI. Start delegating to effective agents.
Leave with proactive agents that do real work for you.

Cohort 5 · kicks off Tuesday 17 November 2026Suitable for non-technical folksAdvanced lane for Claude Code users

700+ people trained so far

Cohort seat £900 +VAT with a 100% money-back guarantee.
Self-paced = every recording, guide, and template, in your own time - £375 +VAT.

Works with your existing AI setup

  • Claude Chat
  • Claude Cowork
  • Claude Code
  • ChatGPT For Work
  • Codex
Week 117 Nov – 20 NovYour second brain, safety & your tools
1
Tue · 17 Nov · 2:30–5:00pm UK

Agent Foundations: Your Personal Second Brain

Build in the app you already use: Claude, Cowork, Claude Code, ChatGPT or Codex

Everything in this programme is built on one asset: your personal second brain, a folder of files about you, your work and your goals that any agent can read. In this first session you build it through a guided interview, open it in the AI app you already use, and have your first real conversations with an agent that knows who you are. Every session after this one builds on it.

What you'll build

  • Your second brain folder, open in the app that fits you: Claude, Cowork, Claude Code, ChatGPT or Codex
  • The first files in your second brain, written for you by a coaching-style interview about your role, your goals and how you work
  • Standing instructions (a CLAUDE.md or AGENTS.md file) your agent reads at the start of every chat, so it stops giving generic advice
  • Your first real chats with it, on work you are doing this week

What you'll learn

  • What an agent actually is: it thinks (the model), knows (your context) and does (the tools it can use)
  • How the pieces fit together, pictured as a restaurant kitchen: the agent is the chef, connectors are the deliveries, skills are the recipes and scheduled tasks are the kitchen opening at 7am
  • Why a second brain beats relying on chat memory or clever prompting, and how standing instructions shape everything your agent does
  • Claude or ChatGPT, and which model to choose for which job

Going further (advanced)

  • Turn your second brain into an LLM wiki: linked pages that get richer every time you add to them
  • Sync it across your devices, so the same brain is with you wherever you work

What you won't get anywhere else

I'll share two prompts I've refined over months of daily use: an Interview Prompt that interrogates your communication style, decision-making, and blind spots to generate deeply personalised instructions, and a Personal OKR Prompt that gets you crystal clear on what you're actually optimising for. One Cohort 3 graduate called building his second brain 'a game changing moment'; another called this session 'the biggest single breakthrough' of the programme.

2
Fri · 20 Nov · 2:30–4:30pm UK

Security, MCPs, APIs & CLIs

Wiring your agent into your real tools, then locking it down

Your agent goes from smart but blind to reading and acting across the tools you use every day. You map how your tools fit together and connect them, then lock it down: tighter permissions on what it can touch, and security rules it follows in every chat. You leave with your first live connections and a clear sense of what is safe to connect and what isn't.

What you'll build

  • A system map: a diagram of your tools and how they connect, drawn by your agent and corrected by you
  • Live connections to your most-used tools, and at least one workflow that runs across them
  • Permissions tightened on at least one tool, so your agent asks before it creates or changes anything
  • A security report on your set-up, and security rules saved where your agent reads them every time

What you'll learn

  • What an API, an MCP server and a CLI are, in plain English, and the five routes from your agent to a tool, starting with the simplest
  • Why to install only official integrations, and why signing in beats pasting an API key
  • Always allow, ask first or block: permissions hold even when an instruction is ignored, and the ladder of trust (read first, then create with your approval, then edit, never delete)
  • Prompt injection and the lethal trifecta: why private data, content from strangers and the ability to send things out should never meet in one agent, and why not to connect your main inbox
  • Onboarding an agent like a new hire: what it gets access to on day one, and what it earns later

Going further (advanced)

  • Give your agent its own email inbox, so it never needs yours
  • Try to trick an agent in the Agent Breaker game, and see prompt injection from the attacker's side
  • Build your own MCP server for a tool that has no connector
  • Install your first CLI, and let your agent drive a browser when there is no other way in

What you won't get anywhere else

We use real moments of confusion from past cohorts (file-sharing permissions, a live prompt-injection demo) as the teaching examples, so the risks land concretely. Then you see the real workflows live: I've run this exact stack daily since late 2025 across consulting, coaching, and sales - a pipeline review that surfaces stale deals, call prep with research pulled in automatically, and post-meeting follow-ups that update the CRM and create tasks in one go.

Week 224 Nov – 27 NovReusable skills & always-on agents
3
Tue · 24 Nov · 2:30–5:00pm UK

Reusable Agent Skills

Named workflows you write once and run forever

Skills are the unlock that makes everything afterwards easier. A skill is a saved recipe: a written procedure your agent follows, so a task you do again and again comes out your way every time. You leave having shipped at least one skill on your real work, with a note of the time it saves.

What you'll build

  • One skill for a task you repeat every week, built with Claude's Skill Creator from real examples of the work done well
  • A rule in your standing instructions so your agent checks any skill before it installs it
  • Your first downloaded skill, installed safely: Humanizer, which strips the tells of AI writing out of a draft

What you'll learn

  • Skills versus freestyle prompting: why a recipe gets you the same quality every time, and asking from scratch does not
  • The anatomy of a skill: its name, the description that tells your agent when to use it, where its data comes from, the output style and worked examples
  • How skills, connectors, standing instructions and plugins fit together, and when to reach for each
  • Where to find ready-made skills, and why a skill you download is someone else's instructions you are about to run
  • Recording a skill from your screen, and why it is a last resort when no connector exists

Going further (advanced)

  • Test your skill with evals: a set of test cases that shows whether a change made it better or worse
  • Let your agent improve your skill one change at a time, keeping only the changes that score better
  • Scan skills with SkillSpector, and add a hook that makes the scan hard to skip

What you won't get anywhere else

We dissect real skills I use every day, then build a brand-new skill live with a build-test-improve loop that compresses days of trial and error into minutes. Newcomers leave with a working skill they can copy; advanced builders leave with the workflow for hardening the skills they already have half-written.

4
Fri · 27 Nov · 2:30–4:30pm UK

Proactive Agents

Scheduled, always-on agents

This is where your agents stop waiting for you. You put one of your skills on a schedule, so a piece of work gets done and lands where you will see it without you asking. You leave with at least one scheduled agent that earns its place (not a daily info dump you learn to ignore), set up so it drafts and never sends.

What you'll build

  • An agent and the skill it runs, tried once by hand with no help from you
  • A one-line scheduled routine that calls that skill, with a run log from day one
  • Safety settings for when you are not watching: it drafts, never sends, and anything risky is switched off
  • A check of its first runs against what you would have done yourself

What you'll learn

  • The briefing trap: why the 200-line morning summary gets ignored by week two, and what to ask for instead
  • The four levels of proactive agents: information, then suggestions, then drafted work, then closing the loop
  • Why skills come first: a routine that calls a skill gives you the same quality on every run
  • Running it on your computer or in the cloud, and why a cheaper model is usually the right one for scheduled work
  • Where the result lands, and the difference between the Slack connector and Claude Tag, a shared agent in your team's channels

Going further (advanced)

  • Close the loop: give your agent a run log it reads back, so it spots what it keeps getting wrong and proposes edits to its own skill
  • Move it to the cloud so it runs with your laptop shut

What you won't get anywhere else

I'll walk through Jamie, my own sales agent, built over weeks of iteration, showing the real output at each of the four levels - from noisy and ignored to trusted with real work - so you can see exactly where yours can go. The skills you built on Tuesday are what make the higher levels possible.

Halfway through the curriculum. ~4.5 hrs of live sessions a week, all recorded. Cohort 5 kicks off Tuesday 17 November 2026.

Book your seat(s)
Week 31 Dec – 4 DecAgent teams & building software
5
Tue · 1 Dec · 2:30–5:00pm UK

Subagents & Adversarial Review

Give your agent a team: helpers that work in parallel, and critics that check the work

One agent doing everything in one long chat fills up and starts forgetting your earlier instructions as the job grows. In this session your agent becomes a manager: it hands the legwork to helper agents (subagents) that each start with a clean desk, and asks a panel of critic agents to look for mistakes in the result before you see it. You leave with one of your skills upgraded to use subagents, or a new skill built around them.

What you'll build

  • A delegation block in your standing instructions, so your agent hands the legwork to cheaper helper agents without being asked
  • One of your skills upgraded to use subagents, or a new skill built around them, with the old and new versions run on the same real work so you can see the time saved

What you'll learn

  • The main agent as a manager and subagents as contractors: briefs go out, short reports come back, and each helper works at its own clean desk
  • When to split the work: keep steps in order when each one needs the one before, and run them in parallel when they are separate pieces combined at the end
  • Choosing the right model for each helper, and the token trap that can use up a month's allowance in minutes
  • Adversarial review: a panel of critics (the buyer, the sceptic, the fact-checker, the editor and a what's-missing critic) that see only the finished piece
  • A second opinion from a different model, such as ChatGPT, for the work that really matters, because one model's critics share its blind spots
  • Keeping actions in the main chat: helpers read, research and draft, and sends and deletes wait for your OK

Going further (advanced)

  • Install the multi-agent-refine skill, which runs the critics, fixes what they find and repeats until the work holds up
  • Set up a cross-model review: Codex checking Claude's work, or Perplexity and Gemini checking each other's research

What you won't get anywhere else

You watch it race: the same competitor analysis run twice side by side, one agent working through five competitors in turn against five helpers working at once. Then five critics take a real proposal apart in under a minute. I use these patterns in my own skills every day, including a weekly review that sends a helper to each of 17 sources at once, so you see how it works in daily use, not only in a demo.

6
Fri · 4 Dec · 2:30–4:30pm UK

Coding Agents

Claude Code, Cowork or Codex (your choice)

A method for building anything more complicated than a single prompt, with you as the product manager: you decide what gets built and why, and a coding agent that already knows how to code builds it. No coding required. You research what exists, write a plan clear enough to hand over, pick a layout, then give it to a fresh agent to build. Software teams ship a working prototype; everyone else ships a validation page to test demand.

What you'll build

  • A research brief on what good looks like and what already exists that you can plug in, so you never build what you could borrow
  • A one-shot PRD (product requirements document), written with the PRD skill as it interviews you: the problem, the decisions, the milestones and what is out of scope
  • Three wireframes side by side, and the one you pick
  • A working prototype or validation page, built from your PRD in a fresh session

What you'll learn

  • The mindset shift: you're not learning to code, you're learning to direct an agent that already can
  • Skill or app: a skill when the job needs judgement every time, an app when it is the same steps every time or other people need to use it
  • The seven-step method: research before building, gather your context, write the PRD, wireframe it, pressure-test it with critics, build it in a fresh session, and review it with annotated screenshots
  • Why the plan lives in a file, not the chat: when a long chat runs out of room the agent forgets, but it can always reread the PRD and pick up where it stopped
  • Which tool fits which job: Claude Code, Codex and Cowork

Going further (advanced)

  • Keep API keys in a .env file, never pasted into the chat
  • Draw your mock-ups in Figma or Paper, connected to your agent
  • A design.md file that keeps everything your agent builds on brand
  • Firecrawl, Perplexity or Claude in Chrome: which web tool to reach for when your build needs data from the internet

What you won't get anywhere else

I build with these agents daily - prototypes, internal tools, landing pages, and full applications, including the platform this course runs on. You get the distilled version: which tool to use when, the prompting patterns that actually work, the mistakes that waste an afternoon, and the shortcuts that save one. Guided building with someone who has done enough of it to know where things break.

Week 48 Dec – 11 DecRoll out to your team & Demo Day
7
Tue · 8 Dec · 2:30–5:00pm UK

Team Rollout: Your Company Second Brain

Growing your personal second brain into a company second brain your whole team draws on

The first six sessions built your agents and your software. This one treats AI as a team sport. The personal second brain from Session 1 becomes the seed of a company one: a shared brain every agent and every colleague can draw on, with an owner, a map and rules that stop it filling up with slop. One Cohort 3 participant built a customer-support brain from 11 internal documents that helps her team resolve tickets when no experienced colleague is online; this session is how you build yours.

What you'll build

  • A map of your team's knowledge: one page that shows where everything lives
  • Your team's rules for the brain, set in a knowledge skill that edits linked pages and pushes back when a change would make the brain worse
  • A plan for your skills library, with a home and an owner for each of your top skills
  • An optional benchmark of where your company is now, from a maturity quiz

What you'll learn

  • The test for what goes in the company brain: would a colleague or a team agent need this to do their job?
  • Where to host it, from Notion to files in GitHub, and why to tidy what you have before you switch tools
  • Page hygiene: an owner, a last-checked date, a summary at the top and links to related pages, on every page
  • Why the brain needs one owner, and a weekly check that catches "slop grenades" (bloated pages nobody reads) before they spread
  • Sharing skills across the team, AI as a teammate in Slack, working with the door open, and hackathons that end in demos

Going further (advanced)

  • Audit your brain with the company-brain-audit skill and get your top fixes
  • Set up a weekly check that runs on a schedule and reports problems without editing a page
  • Turn shared skills into plugins in a team marketplace that syncs to everyone's Claude accounts

What you won't get anywhere else

This is built from watching real companies do it: guest leaders who have opened up their actual company brains inside the programme, from a 200-person scale-up to companies past 1,000 people, plus around twenty company builds from Cohort 3. You see what worked and, just as honestly, what failed - including why exporting files to your team goes stale, and what the companies that cracked it do instead.

8
Fri · 11 Dec · 2:30–4:30pm UK

Demo Day

Demo day

The finale. Members of the cohort demo what they built (demoing is optional, and you choose what to share), we celebrate the standout agents from across the four weeks, and we map out what to build next. In Cohort 2, 24 people demoed working agents four weeks after starting - many of them non-technical when they began.

Book your seat for Cohort 5 - kicks off Tuesday 17 November 2026 →

What's included

4 weeks • 7 live workshops + Demo Day (~4.5 hrs a week, UK time) • weekly guest speakers & department sessions • yours to keep

7× Live Workshops

Tuesdays 2.5 hrs · Fridays 2 hrs · UK time

Each workshop you build something new on top of your second brain, live with the cohort: an hour of teaching, then a guided co-build. Security and integrations, skills, proactive agents, subagents, coding agents and rolling out to your team.

Demo Day

Final Friday · 2 hrs · UK time

The finale: the cohort demos what they built (demoing is optional), and we map out what to build next.

Checkpoints & Certificate

Weekly build checkpoints keep you shipping (with an advanced lane on most), and completing the programme earns a certificate.

Guest Speaker Lunch & Learns

Thursdays

A weekly session with a specialist guest, going deeper on a topic that complements the week. Previous speakers have come from AI-native companies including Fin (Intercom), Incident.io, Jack & Jill and Rig.

Department Sessions

Grouped by department

Drop-in sessions by function to get feedback on your agents and see how others are solving the same problems.

Live Build: An App in an Hour

Optional

James plans and builds an app live in Claude Code, from an idea the cohort votes on, then takes questions.

Q&A Sessions

Before kickoff and after Demo Day

An optional Q&A to get set up before the first workshop, and a final Q&A with James once Demo Day is done.

Private Coaching Calls (Add-on)

Not included - optional extra

Private consulting and coaching calls with James and his coaches are available throughout the programme, on your specific agents and AI rollout.

Private Community

Four weeks inside a group with James and the cohort. Share wins, swap notes, and get unstuck between sessions.

Full Recordings & Guides

Every session is recorded, with step-by-step build guides, so you can catch up or go deeper any time.

Prompts, Templates & Frameworks

The mega-prompts, system instructions, skills, and agent frameworks you build are yours to keep and adapt.

Partner Perks

Discounts on Granola, Attio, and more.

Works with Claude, ChatGPT, and whatever AI tools you already use.

Looking at university courses too? See how executive education AI courses compare.

Frequently asked questions

Do I need to be technical to take the Agent Accelerator?

No. The course is designed to work for non-technical operators: if you've used ChatGPT, you're ready. Every session offers a build container to match your comfort level, with plain-English guides throughout - many graduates began with nothing more than ChatGPT experience.

Comparing it with other courses? See the agentic AI course for people who don’t code.

Can I take the course in ChatGPT or Copilot instead of Claude?

Yes. You can take it in Claude, ChatGPT or Copilot. James teaches it so that nothing is locked into one tool: if you build everything in Claude and then decide to go back to ChatGPT, what you built moves with you.

Will I pay per token for what I use on the course?

Not on a subscription. On a personal or team plan for Claude, ChatGPT or Copilot you pay a fixed monthly fee, and if you hit the plan's usage allowance you're asked to upgrade rather than charged per token. Enterprise plans can be different: Claude Enterprise, for example, bills every token on top of the seat at API rates, so check with whoever runs your company's plan.

Do I work on my own projects or on practice demos?

Your own work, every time. The course never takes you away from your job: there are no side projects or demos, and every exercise speeds up or improves something you're working on right now. The first exercise picks out the pieces of work most likely to move the needle for you, and you put each new agent into practice between sessions. By the end you're closer to your goals at work, and able to take on kinds of work you couldn't before.

How should I prepare for the Agent Accelerator?

Install the Claude, ChatGPT or Copilot desktop app on your laptop and sign in on a paid plan. Then think about your goals at work for the next few months, and start a running list of every task you'd like AI to take on. You don't need anything more: the first session helps everyone work out their goals, so a fuzzy idea is fine. The clearer your goals, though, the sooner your agents are doing work that matters to you.

Does everyone who applies get a place?

No. James checks every application to see whether the course is a good fit. If someone has already covered a lot of what the Agent Accelerator teaches, he points them to his more advanced courses instead.

If that sounds like you, look at the advanced Claude Code course.

How much time does it take each week?

Around 4.5 hours of live sessions a week - Tuesdays 2:30–5:00pm and Fridays 2:30–4:30pm, UK time - plus optional guest talks, drop-in Q&A sessions and office hours, and build time between sessions. Every session is recorded, so a missed week does not derail you.

What do I need before starting?

A paid Claude, ChatGPT or Copilot plan (for Claude or ChatGPT, £20 a month minimum; the free tiers won't get you through the course) and a laptop you can install integrations on. If your company restricts AI tools, there are read-only and no-connector paths through the course, and the Week 1 security session covers working within your organisation's AI policy.

What is a second brain?

A portable folder of context files - who you are, how you work, what you are aiming at - that any AI agent can read. You build a personal one in Session 1, every later session builds on it, and in Week 4 you grow it into a company second brain your whole team can draw on.

How much does the AI agent course cost?

A seat on the live four-week AI agent training (Cohort 5) is £900 +VAT, with a 100% money-back guarantee: complete the programme, attend the sessions (recordings count) and complete the checkpoints, and if you are not satisfied with the value, we refund in full. The self-paced course - every recording, guide, and template - is £375 +VAT. Teams of 10 or more get volume pricing (£900 +VAT per seat).

Is there a certificate?

Yes. Weekly build checkpoints track your progress through the programme, and completing it earns a certificate.

How current is the curriculum?

It is rebuilt for every cohort, and it has to be. The tools change too fast for a recorded course to stay useful for long: MCP, agent skills, scheduled agents and coding agents like Claude Code and Codex all either did not exist or worked very differently a year ago. So before each cohort James goes back through every session and rebuilds it around what people are actually using now, drops anything that has been superseded, and adds the new workflows he is running in his own business that week.

Sessions get replaced between cohorts as the leverage moves, the security session is updated as new attack patterns appear, and because James builds and runs agents daily (including the platform this course runs on), the examples you see are live workflows, not slides from six months ago.

The self-paced course gets the same treatment. When a cohort finishes, its new recordings and guides replace the previous ones on the platform, and everyone with access gets the update for free. Whichever option you choose, you are learning the current version of the craft.

Can I buy the course for my team?

Yes. Teams of 10 or more get volume pricing (£900 +VAT per cohort seat), and we can invoice your company. Email hello@jamesmcaulay.co.uk to set it up.

Ready to build?

4 weeks. 7 live workshops + Demo Day. Agents that take real work off your plate - the follow-ups, the briefings, the busywork - and give you hours back every week.

Cohort 5 kicks off Tuesday 17 November 2026 · seat £900 +VAT · 100% money-back guarantee

Self-paced is every recording, build guide, and template from the programme - £375 +VAT.
Buying for a team of 10+? Book a discovery call.
Rather train the whole team together in a day or two? See hands-on AI workshops for teams.

See all AI agent courses

Questions? Email hello@jamesmcaulay.co.uk