AIQuity Learning
Work the real system, from one model to agents that run together.
A first-principles track for people who need to operate AI, not follow it.
You work the running model, then keep what you can show for it.
By the end, you should be able to:
- 01
Take a running model apart
Tokenizer, context, attention and inference, inspected in the browser rather than described.
- 02
Assemble an agent from its parts
Instructions and context through tools, actions, observation and memory, wired up and run.
- 03
Place a new release without starting over
New models, techniques and frameworks attach to a structure you are already holding.
- 04
Design agentic systems
By reasoning about models, context, tools, orchestration, memory, evaluation and failure modes.
- 05
Show what you worked
Each path ends in a skill contract you check yourself against, and the work you produced stays yours.
Build the whole system in your head before you build it in code.
Every concept has a place.
What a model is made of. What you build around it to make it an agent: context, memory, tools, policy, the loop.
How those agents become systems, and what changes when the systems meet the real world.
AI changes every week. The underlying systems don’t change every week.
New models, tools and frameworks will keep arriving.
Chasing every update doesn’t create a durable mental model.
Learn the layers underneath, and new developments have somewhere to attach.
The machine
LLMs, training, inference, context, serving.
The agent and its harness
What you build around a model so it holds up: context, memory, tools, policies, loops and action.
The system
Orchestration, workers, shared memory, coordination and control.