Choose the job you want AI to help with
Five routes turn a large library into a smaller next step: everyday use, professional leverage, product decisions, hands-on building, or the full map. Pick a job first; let the route choose the theory.
THE QUESTION THIS PAGE ANSWERS
ANSWER FIRSTWhat is the key idea behind “Choose the job you want AI to help with”?
Five routes turn a large library into a smaller next step: everyday use, professional leverage, product decisions, hands-on building, or the full map. Pick a job first; let the route choose the theory.
A route is a constraint, not a commitment. Use this page to reduce choice overload. Start with one task you want to finish this week, then follow only the lessons that explain the decisions behind it.
Write the task in one sentence before choosing a route.
A route that sounds impressive but never meets a real task.
Just want to use AI
No coding, no AI product work — you just want it to actually save you time. You'll finish knowing when it's reliable, when it's making things up, and how to ask so you get something useful.
Start hereUse AI professionally
Still no products and no coding — but you want AI as real productivity. On top of “just use it,” you add how LLMs work and a full Vibe Coding playbook: why it invents, how to set rules, so it stays steady when it works for you.
Start hereBuild AI products
PMs, designers, ops — you need to align with engineers on proposals, judge feasibility, and cost it out. After this, you can defend trade-offs in design review instead of getting waved off with “technically impossible.”
Start hereBuild it yourself
Engineers or heavy users who want to write an Agent, ship it, and run it. Every core lesson, no skips; follow all six milestones on the hands-on track. Leave the three hardcore source-code electives for when you have bandwidth.
Start hereWant it all
No trade-offs — take the three hardcore electives too: Grok Build's Rust source, DeepSeek Harness's TypeScript plugin core, and open-source models' distillation plus local deploy. Finish those three and you can take apart any Coding Agent on the market.
Start hereThe numbers only show how many lessons this suggested route recommends from each Part; they never indicate access. Every chapter and lesson below opens directly.
Why it's cut this way
“Just want to use AI” skips all theory and engineering, and only teaches using AI well: what it's doing, why it invents, how to ask so you get answers, what you can safely hand off. Beyond the beginner FAQ Part, three Harness-core themes stay in — context engineering, Prompt engineering, practical tips; from the collaboration-methods Part, three lessons on setting rules with AI and keeping long chats on track — useful every day even if you never write code.
“Use AI professionally” adds two blocks on top of just-using-it: the full LLM-fundamentals Part — so you know why it invents and where the edges are, and you get judgment; the full collaboration-methods Part — the four-step flow, acceptance criteria, and environment safety that keep AI steady when it works for you. Plus two programming-basics lessons on vocabulary and vectors, so knowledge-base retrieval misses start to make sense. Still skips all engineering implementation and product-design content.
“Build AI products” builds on “go pro” with the full Harness set, design patterns and evaluation, plus cost engineering and the self-test center. Aligning with engineers, judging feasibility, costing it out — that's these pieces. Skip code walkthroughs, long-running Agents, security sandboxing, and the three hardcore electives: the hands-on practicum keeps only the product-side slices (image-gen productization checklist, character consistency, what to do when the model dies); Agent Loop and MCP implementation stay for the build-it-yourself route.
“Build it yourself” takes every core lesson, including the full code walkthroughs in the hands-on practicum and the self-improvement Part. Follow the hands-on track end to end: finish the three-tier “what you can do now” tasks at each chapter end, fill all six milestones, and you'll leave with a working Agent and your own collaboration playbook. The three source-code electives aren't on this path — they read other people's implementations and don't block you from building your own.
“Want it all” adds three electives — 63 lessons — beyond the build-it-yourself route. Grok Build anatomy walks a production Coding Agent's Rust source from entry to tool calling; DeepSeek Harness unpacks the everything-is-a-plugin TypeScript base; open source, distillation & local deploy covers what open-source models actually open, how big models get small, and how to run them on your machine. These three are the hardest — and the biggest differentiator.
Turn “Just want to use AI” into a reusable learning action
“No coding, no AI product work — you just want it to actually save you time.” moves learning beyond “I read it once” toward being able to use the idea in a new situation. What lasts is not a polished summary, but a judgment you can use to notice, predict, and act.
Use outcomes to check understanding
Starting from “Still no products and no coding — but you want AI as real productivity.”, try explaining the idea or completing a small task before looking at an answer. Then separate your own reasoning, what a tool supplied, and what still needs checking.
Remembering steps is not the same as owning the method
Turn “Unsure which track?” into a rule in your own words and try it on a different example. Knowledge starts to transfer when you can explain why the action still fits after the situation changes.
From “Just want to use AI” to “Use AI professionally”
“Just want to use AI” grounds the problem in “No coding, no AI product work — you just want it to actually save you time. You'll finish knowing when it's reliable, when it's making things up, and how to ask so you get something useful”. “Use AI professionally” then moves it toward “Still no products and no coding — but you want AI as real productivity. On top of “just use it,” you add how LLMs work and a full Vibe Coding playbook: why it invents, how to set rules, so it stays steady when…”. Together, they show that the lesson is not just a conclusion to remember, but a claim with conditions.
Carry the judgment into the next situation
When learning a concept, complete a small task before looking at an answer, explain your reasoning, and redo it in a different situation. Transfer is stronger evidence than repetition.
- “Just want to use AI”: No coding, no AI product work — you just want it to actually save you time. You'll finish knowing when it's reliable, when it's making things up, and how to ask so you get something useful
- “Use AI professionally”: Still no products and no coding — but you want AI as real productivity. On top of “just use it,” you add how LLMs work and a full Vibe Coding playbook: why it invents, how to set rules, so it stays steady when…
- “The closing point”: “Use AI professionally” adds two blocks on top of just-using-it: the full LLM-fundamentals Part — so you know why it invents and where the edges are, and you get judgment; the full collaboration-methods Part —…
The final “The closing point” brings the discussion to ““Use AI professionally” adds two blocks on top of just-using-it: the full LLM-fundamentals Part — so you know why it invents and where the edges are, and you get judgment; the full collaboration-methods Part —…”. The useful thing to carry forward is knowing which judgments must be revisited when input, scale, or risk changes.
The useful part was not the number of routes, but being allowed to start from a different goal. I chose the application path and stopped worrying about what I did not need yet.
I kept my first week to four lessons and built one tiny experiment after each. I used to think I had to finish all the theory before starting; now a small model-powered demo is already working.
If my goal is mostly office work and research, should I follow the full foundation route or jump into the workflow chapters? I want the structure, but I only have a few hours each week.
A small “done means…” checklist for each path would make it easier to tell whether I can actually do something, rather than simply having read it.
No discussion on this article yet.