"The course is dying as the unit of learning"

Hardman comes from the academic side of learning science, not from the creator economy. She spent eight years inside Cambridge teaching course design, then moved into industry consulting on corporate L&D programs. Her work sits between two worlds — the researchers who study how adults actually learn, and the practitioners who build the training programs companies actually pay for.

That position matters here. The argument in the April post is not a contrarian take from someone trying to sell a new product. It's a learning scientist saying that the format her industry has built itself around for the past twenty years is the wrong unit of analysis.

The headline of the post puts the thesis plainly. Hardman argues that the course was never chosen as the unit of learning on its merits — it was chosen because, in the pre-AI era, it was the only thing that could be reproduced and delivered at scale. Now that AI makes other formats deliverable at the same cost, the case for the course as a default collapses.

The course has been the default unit of corporate learning — not because it worked, but because nothing else scaled. — Dr Philippa Hardman, April 2026

Her framing is structural. She is not saying that all courses are bad, or that every course should be retired. She is saying that the course-shaped container — content broken into modules, delivered in sequence, completed at the learner's own pace — became the standard because it was operationally convenient, and the field built every metric, every platform, and every business model around that container. When the container changes, so does everything built on top of it.

The old finding: courses are about 10% of how people learn at work

The 70:20:10 model from Lombardo and Eichinger, published in 1996, allocates roughly 70% of workplace learning to on-the-job experience, 20% to interaction with colleagues, and 10% to formal courses and training. The model has been criticised for the precision of its numbers — they came from self-reported surveys, not measurement — but its broad ratio has been replicated repeatedly in subsequent research. The order of magnitude holds: formal training is a small slice.

The transfer problem: only 10–20% of training survives

Baldwin and Ford's 1988 review in Personnel Psychology is the founding text on "transfer of training" — the question of how much formal training actually changes behaviour back on the job. Their finding, replicated in Henao-Calad's 2024 thirty-year review, is consistent: only about 10–20% of what's learned in formal training translates into sustained behaviour change at work. The rest fades within weeks.

The compliance paradox: passing the course doesn't mean following the rule

A 2023 internal Salesforce study followed employees who had passed compliance assessments at 90% or higher. When the same employees were observed in practice, only about one in three was actually following the protocols they had just demonstrated knowledge of. Hardman uses this as her tightest illustration of the gap between course completion and real-world capability — the same person can score perfectly on the test and ignore the rule the next day.

Stop measuring inputs and start measuring outputs. Completion rates don't tell you what changed. — Dr Philippa Hardman, April 2026

The new data: $400 billion spent, skills gaps persist

The Josh Bersin Company's 2026 Global Workforce Intelligence Report puts annual corporate training spend at roughly $400 billion globally. The same report finds that 74% of executives still report persistent skills gaps in their workforces. Bersin's framing in the report is direct: the spend keeps rising, the gaps don't close.

The behavioural signal: when given the choice, people skip the training

In a 2026 evaluation of the UK Department for Work and Pensions Microsoft Copilot pilot, 3,549 staff were given access to the tool with optional formal training. 89% chose to learn the tool through self-directed exploration — opening it, trying things, asking it questions — rather than completing the formal training that was offered. Hardman reads this as a revealed preference: when people have a viable alternative to formal training, they take it.

What's missing from the stack is also the point. There are no lectures. No pre-recorded video modules. No certification pathways. No completion percentages. The unit of measurement shifts from "did the learner finish the content" to "did the learner change what they do at work."

Hardman writes that AI is what makes this stack viable at scale for the first time. Each of the first three components used to require a human in the loop — an instructor reviewing real artifacts, a manager catching mistakes in context, a colleague answering a question. AI can now do meaningful versions of all three at low marginal cost. The human-led 10% is reserved for the things AI genuinely can't do well.

Hardman's argument is about corporate L&D. Her examples are corporate training programs, her numbers come from corporate evaluations, and her stack is described in the language of workplace learning. She is not writing about independent course creators selling courses to a paying audience.

But the underlying observation — that the static recorded course is a structurally weak format for adult learning — doesn't change when the buyer's motivation changes. A learner who pays for your course out of pocket is more motivated than an employee assigned mandatory training. They will try harder, push through more friction, finish more of the modules. What they are not is exempt from the way adult brains actually consolidate skill: through doing the work, getting feedback on it, and seeing examples in context.

This is one of the patterns visible across multiple corners of the creator economy right now. The premium products that are working tend to wrap the content in things AI can't easily reproduce — cohort schedules, live calls, community access, feedback on submitted work, accountability structures. The self-paced video library priced at €97 in Europe is the part of the market that's getting squeezed hardest, not because the content is worse than it used to be, but because the format Hardman is describing as outdated has more competition now — including free competition.

• The original post — "The Course Is Dying as the Unit of Learning", Dr Phil's Newsletter, April 2, 2026.
• Hardman's newsletter — Dr Phil's Newsletter. Weekly publications on AI and learning design. Worth a subscribe if this thesis matters to you.
• The 70:20:10 model — overview and critique of the Lombardo & Eichinger 1996 framework on the Wikipedia entry, including subsequent replications.
• Transfer of training research — the Baldwin & Ford 1988 paper on Personnel Psychology is still the standard reference.
• The Bersin report — The Josh Bersin Company publishes the annual Global Workforce Intelligence Report referenced in the post.