Learn with Phoebe · Applied AI

Learn When AI with Phoebe

Six sessions, 45 minutes each, on the most underrated skill in the AI era: routing. Which work goes to a dashboard, which to classic ML, which genuinely deserves the tokens - and what each choice really costs once evals, monitoring and cleanup enter the price. Built on a live ten-workload backlog with real published prices. Do not use a nuclear weapon to kill a mosquito: the mosquito dies either way, and now you also have a crater to explain.

6sessions
$2,315true cost, unrouted
$416true cost, routed
10/10 answers trusted
45min per session

The six sessions 🎯 · one routing playbook, built across all of them

Bring your team's real backlog. Each session adds one routing rule; by session 6 all five run live against Cadence's ten workloads, and then against yours.

🩺 Session 1 · easy

The triage question

Why "should we use AI" is the wrong first question, the two-question test that routes any workload, and what the 95 percent failure statistic actually measures.

▶ Start here 📊 Session 2 · easy

The almost-free tier

Dashboards, SQL and rules: verified once, then cents forever. The 10 GB question that costs six cents as a query and five figures as tokens.

the BI case 🌲 Session 3 · medium

When classic ML wins

Structured prediction is a trained model's home game: the tabular evidence stack, backtestability, and inference for cents on the middle rung.

the ML case ✨ Session 4 · medium

When GenAI wins

Unstructured language, judgment and generation - where the tokens are the bargain. Plus the cascade: Checkr's 5x and Google's 100x from escalating only the residue.

the AI case 🧮 Session 5 · hands-on

The cost and risk math

Real token math on real prices, why the true cost includes evals, monitoring and cleanup, and the error-cost asymmetry that dwarfs every API bill.

the honest number 🗺️ Session 6 · capstone

The routing playbook

All five rules live against the full backlog: $2,315 to $416 while trust climbs 0/10 to 10/10. Then the one-page tree becomes yours.

🔒 the live simulator
easy - everyone medium hands-on capstone

Your passport 🎫

Clear all three quiz questions in a session and it gets stamped. Stamps live in this browser only.

Choose your path 🗺️

Three doors into the same discipline: the tool is chosen by the work, never the other way round.

💼 I approve the AI budget 1→ 5→ 6 the test, the honest number, then the playbook
🔧 I route my own work every day 2→ 3→ 4 each tier's genuine home ground, one session each
🎯 The whole journey 1→ 2→ 3→ 4→ 5→ 6 from tool-first habit to a routing playbook you can defend

What is real here 🧾

A course about honest costing owes you an unusually precise account of its own numbers.

ClaimStatus
The dashboard answers in the simulatorReal. Computed live in your browser from the embedded 168-day dataset when you route a workload there
The LLM answers in the simulatorA deterministic teaching model of documented failure modes, labelled as such on the widget. It calls no API
Every priceOfficial published rates, dated September 2026: Claude Sonnet 5 at $2/$10 per million tokens, BigQuery on-demand at $6.25 per TiB
The 95 percent pilot statisticMIT Project NANDA, taught with its own author's correction: a learning gap in organizations, not model failure - and not peer-reviewed
The 80 percent project-failure figureRAND cites external estimates; what RAND measured is the ranked root causes, led by leaders misunderstanding the problem
Trees beat deep learning on tabular dataScoped honestly: typical, medium-sized tabular data, per Grinsztajn et al. and the follow-up benchmark literature

The knowledge map 🧠

The whole course at a glance - hover a session to spotlight its concepts, click any node to jump in.