What a training programme can and cannot buy

The question people ask is whether a course is worth it. The more useful question is narrower: which of the things a course sells do you actually lack? A programme that solves the wrong gap is expensive at any price, and one that solves the right gap has no fixed price.

Three things a programme can sell

Structure. A syllabus is someone else's decision about order. That is worth something real: the field has no standard curriculum, so the alternative is assembling one from blog posts and hoping your ordering is right. An ordered path saves the specific tax of learning the wrong thing first.

A reviewer. This is the part that is hardest to replicate alone and is routinely undervalued. Writing an analysis no one reads teaches you to write analyses no one reads. A cohort with a practitioner who marks your work — and says "that is not a scoping decision, that is a wish" — compresses months.

A credential and a network. Both are real and both are weak in exactly the way credentials are weak in a new field: nobody has enough hiring history to trust the certificate yet, and everyone in the network is also looking for a first role.

The three things it cannot sell

A customer. No simulator reproduces the thing the job is made of: a real data source that is wrong, a real sponsor who changes their mind, a real deadline that costs money. The public descriptions of the long-format academies are honest about this — the labs are missions you complete, and completing a mission is not the same as surviving week three on site.

The habit of asking for a number. Every course teaches frameworks; almost none will fail you for saying "improve efficiency" instead of naming a metric with a baseline and a window. That failure is the one that matters most in practice, and it is a habit, not knowledge.

A decision. Hiring for this role leans on evidence of having done it: a system that ran, an error rate you can quote, a customer who would take your call. A certificate is at best an input to a conversation; it is not the conversation.

How to read a programme's claims

Four questions separate a serious programme from a content business.

1. Who built the labs, and did they run deployments? Not "industry experts" — a name and a deployment you can find.
2. What does passing require? Watched videos and a multiple-choice test certify attendance. A defended artefact — a scoping memo, a metric table, a readout you present — certifies something closer to work.
3. What do graduates do eighteen months later, and can that be checked? Any programme can show its first ten placements; ask what the denominator is.
4. What happens if you fail? A programme with no failure mode has no assessment, whatever the rubric says.

The answers are usually more informative than the syllabus.

Where the long formats fit

Two shapes are common in 2026. Long academies — the 32-week structure is the most visible example — trade time for breadth, and suit people retraining from an adjacent field who want the whole arc: model basics, application building, deployment practice. Short cohorts trade breadth for feedback intensity, and suit engineers who already build and need their judgement corrected rather than their skills filled in. Question banks and lab libraries sit underneath both: they are drills, cheap per hour, and their value depends entirely on whether you attempt them closed-book and then find out you were wrong.

None of these is a requirement. The role has no licensure and no accrediting body, so the honest framing is a purchase decision: what does this cost, which of the three things above does it give me, and what evidence would I otherwise have to produce myself?

A cheaper test of the same question

Before paying, try to produce one artefact without the course: take a workflow you know, write the metric definition table for it, and get someone with delivery experience to attack it for thirty minutes. If the critique lands and you can act on it, you may only need a reviewer, which is a much smaller purchase than a curriculum. If you cannot even find the workflow worth attacking, the structured path is probably the better buy — that gap is exactly what structure solves.

What this page is refusing to say

It is refusing to rank programmes, because the ranking would depend on where you are starting, which none of them can tell you and this page cannot see. It is also refusing the reverse claim — that courses are worthless and only the field teaches. The field teaches slowly and without feedback, and most people who learn only from the field learn a narrow slice of it extremely well and everything else not at all.

The test is not "course or no course". It is whether, at the end of the programme, you can produce the six capability areas with evidence rather than vocabulary. How the role is actually entered, and what people did instead of a course, is covered in how people become FDEs.