Boss: assemble the system
A bare model just failed three real requests. Can you equip the right capability for each one so it succeeds?
The idea inside
A request hits a bare model and it fails. Equip retrieval, a tool, or a context trim, sometimes a two-step pipeline, until the assistant goes from FAIL to SUCCESS.
After this lesson
You can assemble a working assistant for a task by equipping a frozen model with the right capabilities (retrieval, tools, context trimming), and chain them into a pipeline when one step isn't enough.
Where it leads
You can assemble useful assistants. Now: how do these systems run in practice, as agents?
Inside this lesson
That's the real lesson stage, paused. Claim your pass to operate it.
See how AI actually works, end to end.
This lesson is one stop on the full arc. Unlock all of it, and keep it for life.
What you get
- The 34-lesson main path, a finishable route from a word to agents
- Goal tracks for using AI at work and building AI features
- Boss labs that make you apply a whole act, not just recognize it
- Spaced recall that brings each idea back before you forget
- Course memory: every term defined, with links to where it first appears
- A shareable capability card when you finish the main path
- Lifetime access on every device, every future lesson included
Not videos to watch. You predict, operate the machine, then prove it. That is why it stays.
99 interactive lessons and challenges. No videos, no code.
Free launch pass: lifetime access, no card needed
New here? The first lessons are free to try. Start with lesson 0.1
What this lesson shows
A request hits a bare model and it fails. Equip retrieval, a tool, or a context trim, sometimes a two-step pipeline, until the assistant goes from FAIL to SUCCESS.
The question it opens with
A bare model just failed three real requests. Can you equip the right capability for each one so it succeeds?
The walkthrough, in the lesson's own words
- You didn't pick one fix. You assembled the right system for each request.
- Assembled. The bare model now succeeds. Send in the next request.
- A request hit a bare model and it failed. Equip the capability that fixes it.
- One capability wasn't enough: retrieval finds the right page, but handing the model the whole document would still get skimmed. Trimming to just the matching line is the second stage. That's the point of a system, you chain capabilities, you don't pick a single switch.
- You turned a bare model into a working assistant, one request at a time.
- A bare model is frozen and stateless (5.1). You make it useful by assembling a system around it: retrieval to ground it in a real source (5.2), a tool to act on the world (5.3), a trim to keep facts from being skimmed (5.4), memory to carry notes across sessions, and the honesty to say so when no capability can reach the answer. The skill is wiring the right pipeline for the request.
- None of these were fixed by a bigger model. Each request failed because the model lacked access (a live number it couldn't fetch, a fact it never read, a section it skimmed past). The capabilities you equipped, retrieval, tools, trimming, hand the same frozen model the missing piece. Real assistants are exactly this: a predictor wrapped in a system that feeds it the right context and lets it act.
- And this bench is the small version. Real systems have more slots: they pick which model to call, add evals that score answers before they ship, and put safety gates around risky actions. That layer is Act 7.
- A colleague says: just paste our whole wiki into the chatbot each time, why bother with retrieval? What do you tell them?
- Cost and rot. You just built the alternative: retrieve only what the question needs, keep the model frozen, and let tools fetch live facts. Pasting everything pays for every token on every message and buries the answer in the pile.
Key takeaway
You assembled a working assistant for each task, not by picking a single fix, but by wiring the right capabilities into a system around a frozen model.
What you can do after this lesson
You can assemble a working assistant for a task by equipping a frozen model with the right capabilities (retrieval, tools, context trimming), and chain them into a pipeline when one step isn't enough.
Where it leads: You can assemble useful assistants. Now: how do these systems run in practice, as agents?
This is the written summary. The lesson itself is interactive: you predict, drag and operate the mechanism above, and the reveal answers you.