The case for cohort-based learning in 2026
The arguments against the format have aged poorly. The arguments for it have aged well. Here is where we think the discipline is going.
Cohort-based courses had their first hype cycle around 2020. The format was new enough that it generated genuine excitement and shallow enough that it generated some genuinely bad courses. By 2023 the term had become slightly embarrassing in some circles, the way most education terms eventually do — credentialing, MOOC, bootcamp, microcredential, each in its turn.
We are five years past that hype cycle now. The format has separated, slowly, into the courses that work and the courses that do not. It is worth a look at why.
What survived
The cohorts that survived share a few features. They are taught by people who do the work, not full-time educators. They are small — small enough that the instructor knows each student’s name and project by week three. They run on a real calendar, with a real start and end, and they evaluate finished work. They make the work harder than the format makes it easy.
None of those features are accidents. Each is in tension with what makes a course economically efficient to run, and each is what makes the alumni outcomes possible. A course taught by a full-time educator who has never operated an LLM system in production is not the same course. A course of two hundred students is not the same course. A course with no calendar is not the same course. We learn this lesson again every quarter by watching new entrants discover it.
The arguments against
The standard arguments against cohort-based learning, in 2026, are mostly stale.
The first is that the content is available for free. This was true in 2020 and is more true now. It is also irrelevant. The bottleneck on learning for working engineers has not been access to content for at least a decade. It has been time, motivation, accountability, and the company of other learners. Cohorts deliver those four things; free content does not.
The second is that AI will absorb the format. We hear this from people who have spent more time with AI than with cohorts. A cohort is not a content-delivery system, and the parts that are content can be supplemented by AI tutoring — usefully. The parts that are peer pressure, peer review, instructor adjudication, and shared schedule are not absorbed by AI in any timeline we can see.
The third is that the format does not scale. This argument is correct and is not actually an objection. The format is not supposed to scale. The economics of a small cohort are good enough to sustain a real business, and the alumni outcomes are good enough to justify the price.
The arguments for
The arguments for the format have aged well. The most consistent piece of feedback we get from alumni — across courses, across cohorts, across the five years we have been doing this — is that the cohort changed how they work, not just what they know. That is, in our view, the only outcome worth aiming for in education, and it is one that is structurally difficult to produce without a cohort.
The second-most-consistent piece of feedback is that the alumni network has continued to matter long after the course ended. Engineers who took the course together have hired each other, advised each other, debugged each other’s systems years later. This was not a goal of the format; it is an emergent property of it. We have learned to lean into it.
Where the discipline is going
We expect the next few years to be quieter than the last few were. The hype cycle has burned through; the courses that should not exist have mostly stopped existing; the courses that should exist are getting better at the things that distinguish them. The honest version of the future, from where we sit, is that cohort-based learning will continue to be a small, high-quality corner of the education market — and that the engineers who pay attention to it will continue to outperform their peers who do not.
Solutech intends to be that corner of the market for the foreseeable future. We will keep capping cohorts, keep paying instructors more than the market rate, keep saying no to bundling and scaling and most of the moves that would make the business bigger and worse. If you have read this far, you are probably the kind of engineer who already understands why.