The apothecary’s ghost: Lessons from the last great automation of expertise, By Olumide Awoyemi
- +The Lifecycle of Transformation
- +Phase 1: The Artisan Era (Pre-Disruption)
- +Phase 2: Early Industrialisation (Disruption Begins)
- +Phase 3: Mature Industrialisation (New Equilibrium)
- +What Happened to the Patients?
In 1850, if you had a fever, you visited an apothecary. They were masters of bespoke creation. They didn’t just sell medicine; they formulated it — grinding willow bark, distilling tinctures, and hand-rolling pills based on individual symptoms. They held the “secret sauce” of chemistry in their hands.
In 1850, if you had a fever, you visited an apothecary.
Then came the steam-powered pill press. Within a few decades, companies like Bayer and Pfizer could mass-produce standardised aspirin. But the transformation wasn’t the overnight extinction story we often imagine. Many apothecaries adapted successfully. Some became the pharmacists who interpreted physicians’ prescriptions and counselled patients on drug interactions — a consultative role that persists till date. Others pivoted to custom compounding for edge cases: hormone therapies, veterinary medicines, or formulations for patients with rare allergies. Still, others became the chemists and formulators inside those same pharmaceutical companies, bringing their expertise to industrial scale.
What changed wasn’t whether skilled practitioners could find work — it was where the value lived. The routine, repeatable work moved to machines. The artisan’s role shifted from making the standard product to handling exceptions, design, and oversight.
As we move through 2026, we’re witnessing a parallel transformation in cognitive work. AI isn’t just a tool; it’s the industrialisation of thought itself. And just as the pharmaceutical revolution reshaped medicine while improving access to life-saving drugs, AI is repositioning knowledge workers while democratising expertise.
The Lifecycle of Transformation
The apothecary-to-pharmacy transition didn’t happen all at once. It moved through distinct phases, and different industries today sit at different points on this curve.
Phase 1: The Artisan Era (Pre-Disruption)
Value resides entirely in individual skill and specialized knowledge. Every output is custom. The maker owns the full process from raw materials to finished product. Apothecaries in 1840 exemplified this: each practitioner maintained proprietary recipes, hand-selected ingredients, and personally formulated every remedy.
In 2026, some knowledge domains still operate here. High-end legal strategy, certain types of investigative journalism, and breakthrough scientific research remain largely artisanal. AI assists but doesn’t yet standardize the core work.
Phase 2: Early Industrialisation (Disruption Begins)
Machines handle the routine center while specialists retreat to the edges. In pharmaceuticals, this was the 1880s-1920s: factories produced common remedies while apothecaries pivoted to consultation and custom work. The transition was bumpy but not apocalyptic. Many practitioners found stable footing in the new landscape.
Software engineering sits squarely in this phase right now. AI writes boilerplate code, handles standard CRUD operations, and generates initial implementations. The engineer’s role is shifting toward system architecture, security review, and integration—ensuring the AI-generated components won’t crash the system or introduce vulnerabilities. Junior developers face the steepest adjustment: the rote practice that once built intuition is increasingly automated.
Phase 3: Mature Industrialisation (New Equilibrium)
The dust settles. The industrial process dominates routine production. Human expertise concentrates in three areas: handling complex edge cases, designing and overseeing the industrial systems themselves, and providing the judgment layer that machines can’t replicate. Modern pharmacy reached this phase by the 1950s.
We can glimpse this future in AI-adjacent fields. Graphic design has partially matured: template-based design is largely automated, while senior designers focus on brand strategy, complex campaigns, and art direction. The profession didn’t disappear—it reorganised around where human judgment remains essential.
Education is often cited as ripe for AI transformation, but the reality in 2026 classrooms is more complex than the thought experiments suggest. AI can provide explanations and even adapt content to some degree, but effective teaching involves much more: diagnostic assessment of misconceptions, motivational scaffolding, managing classroom dynamics, and the social-emotional aspects of learning.
What we are seeing is more subtle. Teachers increasingly use AI to generate initial lesson materials, differentiated practice problems, and assessment items—similar to how pharmaceuticals freed pharmacists from manual pill-rolling so they could focus on patient consultation. The shift is not from teacher-as-knowledge-dispenser to teacher-as-mentor (a false dichotomy—good teachers always did both). Rather, it is from teacher-spends-hours-creating-worksheets to teacher-spends-hours-understanding-individual-student-thinking.
Where this leads remains genuinely uncertain. Early-phase disruption is messy, and education sits earlier on the curve than software engineering. We should be cautious about declaring the endpoint when we’re still watching the transition unfold.
The most severe casualty of industrial drug production was the apprentice role. When factories eliminated the need for someone to manually grind compounds and roll pills, they eliminated the primary path to expertise. The master apothecary learned medicine’s fundamentals through years of physical practice before advancing to formulation. Remove the practice, and you risk creating a generation that knows which buttons to press but not why the machine works.
We’re encountering this same pattern in 2026. If AI handles all entry-level drafting, legal research, and code review, where do tomorrow’s senior attorneys and principal engineers develop their instincts? Mastery doesn’t come from reading about the work—it comes from doing the work badly at first, building pattern recognition through repetition, and gradually internalizing the principles.
This is the junior gap, and it’s not hypothetical. Companies are already noticing that AI-assisted junior developers struggle when the AI fails or produces subtly broken code. They haven’t ground enough herbs to recognize when the mixture doesn’t smell right.
The solution isn’t to reject automation—that ship has sailed. Instead, we need to deliberately architect apprenticeship into the AI era. This might mean structured manual-mode phases where juniors work without AI assistance, explicit rotation through different types of problems to build breadth, or pairing junior workers with senior mentors who can translate AI outputs into learning opportunities rather than just accepting them. The pharmaceutical industry eventually solved this through formal education and structured residencies. We’ll need equivalent innovations for knowledge work.
What Happened to the Patients?
