How this whole thing works

What this page is. Most of the course is hands-on — install this, click that. This page is the why. Read it once, the rest makes sense. It’s short.

There’s a bit of setup involved in getting AI to actually pay off for your work. If you understand why you’re doing each step, you’ll stick with it past the awkward first week. If you don’t, you’ll quit when something doesn’t work. So this page first, then the installs.

The problem we’re fixing

If you’ve used ChatGPT or Claude in a browser before, you’ve probably noticed: every conversation starts from zero. Yesterday’s three-hour working session leaves nothing behind except a chat window you’ll never scroll back to. The AI doesn’t remember your name, your projects, your team, or that “Tofo” isn’t a typo. You re-explain every session. The AI never really gets smarter about your work, because each conversation is its own little island.

The setup this course teaches fixes that. Three ideas, layered.

Idea 1 — Externalise the memory

The first move is to give the AI files it can read and edit, and let those files live on your computer.

Your notes — meeting transcripts, project pages, drafts, daily notes — live in a folder of plain markdown files. You edit them in Obsidian (a free, nice editor for those files). Claude Code can read and edit the same files. Same folder, two readers.

What that buys you: every Claude conversation has access to your accumulated knowledge. Yesterday’s meeting transcript is just a file in the vault; tomorrow’s session can read it without you re-explaining. Three months in, when you’ve got a hundred meeting transcripts and a thousand notes, Claude can pull context from any of them. The AI gets smarter about your specific work over time. Not generic-anyone — you.

Think of it as building a personalised version of general intelligence — the smarter the AI’s picture of your world, the more useful its help.

Idea 2 — Two AIs are better than one

Different AI systems have different strengths. The setup uses two:

Claude Code (Anthropic) is the primary. It’s great at file editing, agentic multi-step tasks, and code/analysis work. It’s the one with access to your notes folder. Day-to-day this is where you live.
ChatGPT (OpenAI) is the companion, used for two specific reasons. First, its voice input is excellent — much faster than typing for anything long. Ramble at ChatGPT for thirty seconds, get a clean transcript, paste into Claude Code. Second, Codex (OpenAI’s coding agent) is a useful second opinion on important code or analyses — different AI, different blind spots. Two-AI cross-checking beats trusting either one alone on something that matters.

Why use just one great AI when you could be using two — for the things each is best at?

Idea 3 — Shared organisational intelligence via GitHub

The first two ideas make AI useful for you. The third makes it useful for the whole team.

Beyond your personal vault, MMF has knowledge worth pooling — things any one of us has figured out the hard way that the rest of us shouldn’t have to. A few concrete examples:

How to organise a CITES export permit.
How to program a satellite tag for a particular deployment.
A good workflow for analysing Sharkbook.ai data.
A scarring-rate protocol that’s been validated across three sites.
A grant-report skeleton that’s already worked for a particular funder.

These don’t belong only in one person’s head, or only in one person’s vault. They belong in a place every MMF teammate’s AI workflow can pull from. That place is the MMF organisational GitHub.

The pattern is straightforward:

You read from GitHub. When you need to do something that someone else has already worked out, you pull the relevant note / workflow / skill into your vault and use it.
You push to GitHub. When you’ve worked out a useful pattern, you push the polished version back. Future-you and the rest of the team don’t reinvent it.

The actual mechanics of the push/pull (a “lesson” on GitHub itself) will land here later, once we’ve got a sharper sense of what’s working in practice. For now, just know that the pattern exists and is the missing piece between “AI works for me personally” and “AI makes the whole team smarter.”

The diagram

Personal level — your setup

                   ┌───────────────────────────────┐
                   │   YOUR VAULT  (folder of .md) │
                   │                               │
                   │   notes · transcripts ·       │
                   │   projects · drafts · …       │
                   └───────────────┬───────────────┘
                                   │
                       read & edit the same files
                                   │
                    ┌──────────────┴──────────────┐
                    │                             │
              ┌─────▼──────┐               ┌──────▼──────┐
              │  OBSIDIAN  │               │ CLAUDE CODE │
              │   (you)    │               │    (AI)     │
              └────────────┘               └─────────────┘

                  Plus ChatGPT (voice input)
                  and Codex (second-opinion review)
                  for the things each is best at.

Org level — the team

       ┌──────────────────────────────────────────────┐
       │     MMF GITHUB  (shared knowledge base)      │
       │                                              │
       │   - reusable workflows & skills              │
       │   - team how-tos (CITES permits, satellite   │
       │     tagging, Sharkbook analyses, etc.)       │
       │   - shared templates                         │
       └────────┬──────────────────┬──────────────────┘
                │                  │
                │   pull when      │   push back when
                │   needed         │   improved
                │                  │
   ┌────────────┴───┐   ┌──────────┴────┐   ┌──────────────┐
   │  Your setup    │   │  Teammate's   │   │  Teammate's  │
   │  (Claude +     │   │  setup        │   │  setup       │
   │   Obsidian)    │   │  (same shape) │   │  (same shape)│
   └────────────────┘   └───────────────┘   └──────────────┘

Same individual pattern, replicated across the team, with GitHub as the connecting tissue. One person figures something out; everyone benefits.

You don’t build all of this on day one

If this is starting to sound like a lot, it isn’t — not on day one. The progression is:

Install Claude Code and Obsidian on your Mac (Lesson 2).
Point Claude at your notes folder (Lesson 3).
Let Claude interview you for fifteen minutes about how you work (Lesson 4).
Try one real workflow — transcribe a meeting (Lesson 5).
Adopt the daily rhythm (Lessons 6–7).

That’s the universal on-ramp. Most people stop there for the first month and just use it on real work. The GitHub-sharing layer is the next ring out; specialised AI workflows (research, scientific writing, citation checking) the ring beyond that. Pick up each ring when you want it — there’s no compulsory order beyond the first five lessons.

What’s next

Install the software you’ll need — two free apps, about ten minutes. Once those are running we can connect Claude to your notes folder and you’ll have an actual working setup in maybe twenty minutes total.