saikodev.ai

Start here

The AI learning roadmap

Everyone asks the same question: where do I start? This is the answer I actually give. Four stages, from never-touched-it to shipping for other people — find the stage whose finish line you have not crossed yet, and start there. You do not need to be a developer to walk most of this path.

  1. STAGE 0Complete beginners — anyone who has only watched AI from a distance

    See what it actually is

    Build first-hand intuition for what these models are good and bad at. No code, no math.

    • Use one frontier assistant every day for two weeks, on real tasks — drafting, summarizing, planning, explaining something you half-understand. Intuition cannot be read into you; two weeks of daily use beats any explainer.
    • Learn the eight or so words that matter — model, token, context window, prompt, hallucination, fine-tuning, agent, eval. That vocabulary is enough to read everything else on this site.
    • Watch how practitioners actually think. The Notes section summarizes interviews and talks with the takeaways written out — start with the Fireworks interview and its point about what work AI is quietly absorbing.

    You're done whenBefore you ask, you can usually predict whether the model will handle a task well — and you are right most of the time.

  2. STAGE 1Product managers, professionals, small-business owners — still no code

    Put it to work

    Move from chatting to a repeatable workflow that saves you real hours.

    • Pick one task you repeat every week and build a reusable prompt for it. Iterate until the output is usable without edits. One systematized task teaches more than twenty one-off conversations.
    • Know where your data goes before you paste anything private — training toggles, retention windows, and the fine print differ sharply between tools and plans. The data-policy comparison covers the four big ones.
    • Tour the tools worth your time, then pick one and go deep. Tool-hopping feels like progress and teaches nothing.

    You're done whenAI saves you hours every week on work you actually had to do — and you can explain to a colleague specifically how.

  3. STAGE 2Developers new to AI, and ambitious non-developers

    Build your first thing

    Ship one small working AI application, end to end.

    • Pick a coding agent and let it teach you — Claude or Cursor are the two I reach for. You describe, it builds, you read what it built. That loop is the fastest programming education that has ever existed.
    • Start from a starter, not a blank repo. The Starter Kits section is being stocked with clone-and-go kits where the boring parts — auth, evals, deployment — are already decided.
    • Build for yourself first. A small tool you actually use every day beats a grand idea for an imagined audience — it keeps you honest about whether the thing works.
    • Optional, for the curious: how coding agents work underneath — the DeepSeek Harness deep dive takes one apart.

    You're done whenSomething you built runs, and you have used it yourself three days in a row.

  4. STAGE 3Side-hustle builders, engineers going deep, non-profits

    Ship it to other people

    Put something real in front of users — and know it works from evidence, not vibes.

    • Write evals before you scale. Most failed AI projects die of vibe-testing, not model choice — and people who can write evals are scarce. The Fireworks interview notes make the industry case for why this one skill pays for itself.
    • Study whole projects, not snippets. Builds documents projects end to end — the architecture calls, what broke, and the repo.
    • Pick a niche you personally know. A small team serving one vertical well beats a small team building a general tool — the specific knowledge is the moat, not the model.
    • Building for a non-profit, or are one? I build AI systems for non-profits free.

    You're done whenSomeone you have never met uses your thing — and your evals, not your feelings, tell you it works.

This roadmap grows as the site does — stages pick up new links as builds, starters, and write-ups ship. If something here is unclear at your level, that is a bug in the roadmap, not in you: tell me.

Or just start reading