Direction. Advanced

What this is

A four-week technical cohort for people already shipping with LLMs who want to go from interesting or experimental to irreplaceable infrastructure. You'll learn how to govern a system of agents and workflows so it doesn't quietly fail six months after you ship it

Who fills these seats

Already shipping

You've built and deployed AI-augmented apps and the one-off agents are starting to accumulate. You want an architecture that doesn't require you to hold it all in your head.

Inheriting the mess

You walked into someone else's autonomous-agent disaster. You need to figure out what's load-bearing, what's cosmetic, and what's been silently failing for months.

Five engineers. One million lines of code.

Under-resourced engineering org with Claude credits. You have to do more than your headcount allows and can't afford to automate the wrong things first.

Converting to AI-native

You're leading a company toward AI-native operations and need an operating model that's honest about what AI should drive, what it shouldn't touch, and what governance sits between them.

I am one of those →

The four questions we answer

Every week is structured around your actual situation — not a case study.

01. What can we concretely measure what is and is not working?
02. How do we instrument what we want to change?
03. How do we pick our tools and how are they implemented?
04. Maintining organizational Alignment

What you produce

Week 1 — Map

A written diagnostic of your actual situation

Week 2 — Instrument

Working code or workflow that logs the thing you most need to see, with sample output

Week 3 — Decide

A working build plus a test suite, a tradeoff doc, and a one-page spec of what it can do autonomously and what it can't

Week 4 — Align

A coherence document: what has to be true for this to count as still working, and how you'd know if it wasn't

What you leave with

— A working diagnostic on your own organization. Complete through all four questions on your actual situation. No need for an outside consultant.

— A coherence document specific enough that someone else could detect drift in your system without asking you.

— Three written critiques of peer work. The diagnostic instinct develops when you apply it to someone else's situation.

— The Direction Advanced Technical Reference Regularly updated best in class reference for the state of art implementations pulled from real world examples

The terms

$1,000 flat. Class every Thursday at 8pm ET.

You're probably spending more than this on wasted credits every month.

The Specific Technical Skills You Will Learn

Infrastructure

Token economics

Context window management

Self-hosting

Prompt caching

JSON mode

Tool use

Observability

Cost per call attribution

Logging schemas

Structured output

Routing + fallback chains

SQLite for logs

Evals + Testing

Golden datasets

LLM-as-judge

Automated regression

TDD with agents

CI eval loops

Tautology trap

Alignment

Four shapes of drift

Coherence criteria

Model version pinning

Weekly eval cadence

Lifecycle + retirement

Ok that makes sense →