Course: agentic financial workflow engineering with Elixir, Ash and AshAI

From cognokratos/tauros-revenue · docs/LEARNING-PATH.md · pinned revision facbbc927eb4

One question runs through the whole course:

How do you let an AI take part in a financial workflow without giving it financial authority?

Tauros answers it with architecture, one layer at a time. Each lesson adds one layer and lets you attack it:

identity → ownership → immutable financial intent → lifecycle constraints
→ idempotency → human authority → exact approval → concurrency → audit
→ narrow AI capability

By the end you can point at the exact lines that stop an AI from approving an invoice, through the UI, REST, MCP or a prompt it was tricked into following.

Who it is for

Software engineers interested in agentic systems, financial workflows, safe authorization, Ash or MCP. You do not need to know Ash; you should be able to read Elixir. This is not a Phoenix tutorial: each lesson is about an architectural decision and the code that enforces it.

Setup (once)

docker run -d --name tauros-postgres -e POSTGRES_PASSWORD=postgres -p 5432:5432 postgres:17-alpine
mix setup        # deps, database, demo data: an approver, an agent, proposals to review
mix phx.server   # http://localhost:4000, sign in as demo@tauros.local / tauros-demo-password
mix test         # every lesson's guarantees, as tests

For the console labs, paste the setup block at the top of EXERCISES.md into iex -S mix. For the MCP lessons you also need curl and jq.

How a lesson works

Every lesson has the same rhythm: Goal · Concept · Code to inspect · Run it · Break it · Why it fails · What to remember · Next. You read a little, run a test or the app, attack the guarantee, and then find the guard that stopped you. The deep explanations live in concepts/; the runnable labs in EXERCISES.md; the course links to them instead of repeating them.

The course

Part I · Identity and ownership

Who may act, and on what?

Part II · Financial intent

What exactly is being proposed, and how does it move?

Part III · Human authority

Who decides, on exactly what, and can we prove it later?

Before Part III: run mix setup so there are proposals to review.

Part IV · AI capability

How do we let a model in without letting authority out?

Before Part IV: finish Part III. You should be able to name the policy that stops an agent from approving before you give an AI a way in.

The answer, in one place

When you finish, compare your answer with AI-AUTHORITY.md · What exactly stops an agent from approving an invoice?

Reference while you learn