A twelve-week AI bootcamp is enough to change careers if you already work in something adjacent — analysis, QA, web development, IT support, a technical sales or ops role — and you use the twelve weeks to build and deploy, not to watch. It is not enough on its own if you have never programmed, and any bootcamp that tells a complete beginner otherwise is selling the calendar rather than the outcome. Twelve weeks at fifteen hours a week is 180 hours. That is a lot of building. It is not a lot of learning to think in code from nothing.
We run twelve-week bootcamps, so read this as the case we would make to someone deciding whether to join one — including the cases where the answer is "not yet".
What can 180 hours actually produce?
Six deployed projects, each with a live URL and a test set, if the programme is built that way. That is the unit that matters. In our count of 315 AI job ads from July to September 2026, production or shipped appeared in 39% of ads — more than any tool — and portfolio or GitHub in 17%, against a degree in 12%. Six things you can show, each of which you can explain, is above the bar that most of those ads set.
What 180 hours cannot produce is depth in the theory, fluency in a second language, or the instincts that come from a year of production incidents. Those come after the first job, on the job. The bootcamp's task is to get you to the first job.
Who changes careers successfully in twelve weeks?
People with three things already: they can make a computer do something and fix it when it does not; they have a specific role in mind rather than "AI"; and they can put fifteen hours a week in without heroics. The profiles that fit are recognisable: a data analyst who writes SQL and some Python moving into AI engineering; a QA engineer moving into LLM evaluation, a job that barely existed two years ago and now appears in a quarter of the ads; a support-team lead moving into a forward-deployed role because they know what customers actually do with software.
The common thread is adjacency. A bootcamp does not teach these people to work; it teaches them the specific stack the new role uses and gives them the proof. We are a new school and do not yet have a placement rate to quote — anyone who does quote one, ask how it was measured.
Who should not start a bootcamp yet?
Anyone who cannot write a hundred-line program in one language without looking everything up. A bootcamp assumes that skill and builds on it from day one; without it, week two is a wall, and the weeks after it are spent catching up on fundamentals while the cohort moves on. That person is better served by eight to twelve weeks on Python for AI first — a smaller, cheaper commitment that turns the bootcamp from a struggle into a sprint. The free skill checks exist to tell you honestly which side of that line you are on; they take three minutes with a free student account.
The other person who should wait is the one who cannot protect the hours. Fifteen a week is the floor. A programme done at seven hours a week takes twice as long, and the projects half-finished at the end are worth less than three finished ones would have been.
Bootcamp, self-paced or a degree?
It depends on what you are buying. A bootcamp buys structure, deadlines, a cohort and graded feedback — the things that make people finish. Self-paced buys flexibility and costs discipline; the completion rates for self-paced online courses have sat below fifteen percent for a decade, not because the material is bad but because nothing is waiting for the work. A degree buys theory, credentials and time, and is the right choice for research roles and the doors that require it. Bootcamp vs self-paced vs AI-graded learning sets the three out side by side; how long it takes to learn machine learning well enough to get hired gives the timelines.
What should a twelve-week AI bootcamp contain in 2026?
The stack the ads are asking for, built rather than described. A programme worth its fee in 2026 should have you deploy, in roughly this order: a model-backed API with a test set; a retrieval (RAG) system over a real corpus; an agent that uses tools; an evaluation harness that produces a number and a list of failures; something that runs in a container with a deploy pipeline and monitoring; and a capstone that combines them for a user who is not you. That list is the top of the job-ads data — agents 39%, evals 23%, RAG 11%, Docker and CI/CD 10% each, monitoring 17% — in the order you can learn it.
If a bootcamp's syllabus is mostly lectures on how transformers work, or if there is no deployed artefact at the end of each block, it is a course with a cohort, not a bootcamp.
Square 1's bootcamps are twelve weeks, fifteen hours a week, six two-week blocks, one deployed project per block, a gate between blocks that you must pass to continue, and grading against a published rubric by Nova, the platform's AI tutor. The Agentic AI, LLM Evals and Reliability and ML Engineer bootcamps are the three most people start with; the full list is on the courses page.
What happens in the twelve weeks after the bootcamp?
Interviewing, and rebuilding the weakest project — and this is where most of the career change actually happens. Plan for a three-month search after a three-month programme. The projects are the interview material; the eval numbers are the answers to "how do you know it works"; the incident write-up is the answer to "tell me about something that went wrong". How to build an AI portfolio hiring managers actually read is about presenting the work; the hiring board lists employers who read graded records instead of CVs.
The honest scorecard
- Adjacent role, can code, fifteen hours a week: twelve weeks is enough. Expect to be interviewing within a month of finishing.
- Adjacent role, can code, seven hours a week: it is enough, but it is six months, not three. Plan for that rather than pretending.
- No programming yet: do the Python course first. Then the answer is yes.
- Want a research or ML-scientist role: a bootcamp is a good first step and a degree is the second one. Twelve weeks alone will not get you there, and nobody honest will say it will.