Hours

The course is sized at 200 guided hours (COURSE-SPEC.md). The table allocates that budget per module and sums exactly. At about 10 hours a week that is 20 weeks; full-time learners can halve the calendar, not the hours. The guided hours are the instruction, labs, targets and the first build of each project. The real-user periods, the job search and the track depth after the core are scoped in the modules but not counted, because the evidence says the whole transition takes 18 months to 3 years and no honest budget compresses that.

The 200-hour path

Stage	Module	Hours	Labs and targets inside
1	01 Software fundamentals for AI systems	10	run_all
1	02 Math and classical ML as table stakes	10	01
2	03 Tokens and the next-token objective	6	05, 07
2	04 Attention and the Transformer	14	02
2	05 Training, scaling laws and inference	13	03, 06, 08; target stale cache
2	06 Post-training: SFT, preference and RL	12	09
3	07 Retrieval done properly	13	10, 04
3	08 Evals as a discipline	16	04, 13; target leaky harness
3	09 Agents and context engineering	13	04, 11
3	10 Cost, latency and security	9	10, 11, 13; target leaky agent
3	11 Fine-tuning for behaviour	8	09
4	12 The internal move and public work	10
4	13 Interviews the 2026 way	8
5	14, 15 or 16: one track	24	by track
6	17 Expert markers and staying current	4	13
6	Capstone	28
6	Final scenario exam	2
	Total	200

Milestones, the final exam and the diagnostic are counted inside the stages they close. The optional Foundation bridge adds 40 hours in front; the Path A on-ramp adds about 210 (A1 to A5) for readers starting from zero, who then skip Stage 1; A6 to A8 are optional.

How the numbers were made

The 500-hour first allocation took the midpoints of the practitioner ranges in the 2026 roadmap research. The 200-hour budget scales each module by the same factor and rounds, keeping the proportions the evidence supported (the model from the inside and the measurement stage carry more than half of the hours).