Where to start, and what it is honestly like.

The free options worth knowing are the System Design Primer on GitHub, which is a reference rather than a course, ByteByteGo's free YouTube material, and engineering blogs from companies operating at scale. Sourcemap covers system design across days 47 to 66 as part of a sequenced ninety-day path, free, which is useful if the problem is that a reference does not tell you what to do on a Tuesday. Most paid alternatives are interview preparation rather than design instruction, and the difference matters more than the price.

Several, and they are genuinely good. freeCodeCamp grades your projects automatically and issues free certifications. The Odin Project is a full-stack curriculum with substantial projects and an active Discord. CS50 is free to audit and its problem sets are graded. Sourcemap is free and structured as ninety fixed days that carry on past web development into production engineering, system design and AI engineering. What no free option replaces is the accountability of someone noticing when you stop.

Six to twelve months of consistent part-time work to reach employable, and considerably less to reach useful. The variable that decides it is not talent, it is whether the work is sequenced. People who study for a year without an order usually cover less than people who study for three months with one. Sourcemap allocates days 35 to 46 to backend and data specifically, inside a ninety-day path at two to three hours a day, which is 214 hours in total.

Yes, and for most jobs called AI engineering you should. AI engineering is building systems on top of models that already exist: prompting, retrieval, embeddings, evaluation, cost, latency, tool calling and agents. Machine learning is training models, and it needs linear algebra, calculus and statistics. They are different jobs. Sourcemap covers days 67 to 90 as AI engineering and states explicitly that it does not teach machine learning.

The gap most people hit after a project-based web curriculum is not another framework. It is the reasoning underneath: why this database, what breaks at ten times the traffic, what to cache, how to tell whether a system is healthy. That is production engineering and system design. Reading the System Design Primer, following engineering blogs, and working through a sequenced path like Sourcemap's days 47 to 66 all serve that, and all of them are free.

Tutorial hell is a retention problem wearing a motivation costume. Following along feels like understanding, and understanding fades within days if it is never retrieved. Three things break it: build something small without a tutorial open, answer questions from memory before revealing the answer, and revisit weak topics days later rather than moving on. Any structure that forces recall works. Sourcemap builds all three into each day and schedules weak concepts to return at one, three, seven, fourteen and thirty days.

Yes. roadmap.sh gives you the map without the sequence. The Odin Project and freeCodeCamp give you sequenced project work in web development. CS50 gives you graded computer science foundations. Sourcemap gives you ninety fixed days across software engineering, system design and AI engineering with a daily time budget, spaced review, and nothing behind a payment. Picking between them mostly depends on whether your gap is knowing what to learn or actually doing it.

Most roadmaps say nine to twelve months for somebody who already programs, and fifteen to eighteen for a career changer starting from no code. Those numbers assume you are also assembling the curriculum as you go, which is where most of the time actually goes. The work itself is smaller than it looks: prompting, retrieval, evaluation, cost and tool calling are a few hundred hours, not a few thousand. Sourcemap allocates days 67 to 90 to it inside a fixed ninety-day path.

No, and conflating the two is the most expensive wrong turn in this field. Machine learning is training models, and it needs linear algebra, calculus and statistics. AI engineering is building systems on models that already exist: prompting, retrieval, embeddings, evaluation, cost, latency, tool calling and agents. They are different jobs with different maths, and almost every role advertised as AI engineering is the second one.

The judgement AI cannot supply. Generating a working function is now cheap, and the scarce skill is reading what came back: knowing what breaks first under load, why this database and not that one, where the security hole is, and when a confident answer is wrong. That is architecture, tradeoffs and debugging, and it is the entire premise Sourcemap is built on.

Yes, and most working developers partly are. The pattern that separates the ones who make it is not talent, it is that they follow somebody else's curriculum instead of assembling one from scattered tutorials. Deciding what to learn next, every night, is the tax that ends most self-taught attempts, and a fixed sequence removes it.

Eight to twelve weeks of consistent study gets you to the point of reasoning about a system rather than reciting a diagram, which is the threshold that matters. What extends it is starting with the diagrams instead of the numbers: capacity, latency and failure modes are what the diagrams are answers to. Sourcemap gives it days 47 to 66, after the production engineering it depends on.

In that order, and close together, because AI systems are systems first. Retrieval is a database question, agents are a reliability question, and the cost of a feature is a capacity question. Studying them separately is how people end up able to call an API but not to say what happens when it is slow. Sourcemap sequences them deliberately: system design across days 57 to 66, then AI engineering from day 67, with the earlier reasoning assumed rather than repeated.