EPHMRA Senior Directors Forum — September 2026

The frontier is here. Your organisation isn’t.

A guided reading room to support Duncan Arbour's viewpoint and provocations from September 23rd.

Start here

Five readings that carry the argument from historical precedent to immediate action.

Foundational paper Paul A. David · 1990

The Dynamo and the Computer

Why installing powerful technology is not the same as reorganising around it.

Subtitled “An Historical Perspective on the Modern Productivity Paradox”, Paul David’s seminal examination of the nature and level of change required to successfully derive value from a General Purpose Technology—in this case, electricity—has been doing the rounds among AI thought leaders for the past year.

For Insights: new tools create local efficiency; redesigned work creates enterprise value.

Read the paper

Go deeper

The Productivity J-Curve — a later formal economic treatment Exponential View #602 — the present-day context

Expert essay Ethan Mollick · 2026

The Overhang

Deep knowledge, wide knowledge, taste and agency are not consolation prizes.

Ethan Mollick is an associate professor at Wharton, an artificial intelligence researcher and the author of the New York Times bestseller Co-Intelligence. His Substack is essential reading for non-technical people trying to understand the impact of AI on knowledge work.

For Insights: do not compete with AI on output; use expertise to direct work neither human nor machine could do alone.

Read the essay

Supporting research

Agentic coding and persistent returns to expertise — Anthropic, 2026

Working paper Catalini, Hui & Wu · 2026

Some Simple Economics of AGI

Execution is becoming abundant. Verification is not.

Christian Catalini—founder of the MIT Cryptoeconomics Lab and a Research Scientist at MIT Sloan—became a cult figure in the AI community earlier this year with this unified economic theory of the AGI transition. If detailed economic theory isn’t your thing, don’t try to read it all: just give it to an LLM and ask for a summary and all the best soundbites.

For Insights: protect the route to expertise while building verification capacity that can scale.

Read the working paper

Preprint Peng et al. · revised 2026

Digital Twins as Funhouse Mirrors

Five distortions behind plausible but unreliable synthetic people.

Worth it for the title alone: a very Westworld perspective on “digital twins” that present as too smooth, stereotyped and hyper-rational compared with their human models.

For Insights: synthetic stakeholders can help generate hypotheses; they are not yet substitutes for decision-grade human evidence.

Explore the study

Peer reviewed Dell’Acqua et al. · 2026

Navigating the Jagged Technological Frontier

AI can improve knowledge work—and make confident people wrong.

AI can’t do everything with the same level of quality and confidence. Some knowledge domains are deeply understood because of the volume of information available in models’ training data; others remain more obscure. It is a foundational paper in the field and introduces a key concept to understand.

For Insights: expertise is how we recognise where AI succeeds, where it fails and when the difference matters.

Read the research

Follow the three provocations

Open a question to see the evidence most useful for taking it back into your organisation.

1 Spend the efficiency dividend on expansion Use cheaper execution to make yesterday’s unaffordable ambition routine.

Begin with the Jagged Frontier.

Peer reviewed Brynjolfsson, Li & Raymond · 2025

Generative AI at Work

A study of more than 5,000 customer-support workers found higher productivity, with the largest gains among less-experienced workers. AI helped transmit the tacit practices of stronger performers.

Read the paper

2 Redesign the team—not merely the toolkit Move from operating software to directing human and machine work.

Begin with David, Catalini and Funhouse Mirrors.

Peer reviewed Vaccaro, Almaatouq & Malone · 2024

When combinations of humans and AI are useful

Across 106 experiments, human–AI combinations beat humans alone on average—but not the better of human or AI alone. Synergy depends on task and division of labour.

Read the meta-analysis

Peer reviewed Bisbee et al. · 2024

Synthetic Replacements for Human Survey Data?

Synthetic responses can reproduce plausible averages while failing to preserve human variation and the relationships between variables needed for inference.

Read the paper

Peer reviewed Jack, Cooper & Flower · 2026

Automating the qualitative interview?

An empirical examination of simultaneous, adaptive chatbot interviews—and why scale must be accompanied by methodological oversight and critical reflexivity.

Read the paper

3 Rebuild the route to expertise Make cognitive offload support mastery rather than replace it.

Begin with The Overhang and the Missing Junior Loop.

Peer reviewed Lee et al. · CHI 2025

The Impact of Generative AI on Critical Thinking

In 936 examples from knowledge workers, confidence in AI was associated with less critical thinking. The cognitive work shifts towards verification, integration and stewardship.

Read the paper

Peer reviewed Bastani et al. · 2025

Generative AI without guardrails can harm learning

Unguarded AI improved assisted performance but left students performing worse when the tool was removed. A constrained AI tutor avoided the same damage.

Read the paper

Research report Hitzig et al. · Anthropic, 2026

Agentic coding and persistent returns to expertise

Analysis of roughly 400,000 agentic coding sessions finds that greater domain expertise leads both to more successful work and to more work delegated per instruction.

Open the report

Life sciences in 2026

Industry perspectives—not peer-reviewed research—showing what leadership teams are being told now.

Industry perspective Deloitte · June 2026

Confidence under pressure

AI deployment is advancing faster than measurable value.

Based on an April 2026 survey of 150 senior life-sciences executives and analysis of first-quarter earnings calls. It brings the capability-overhang argument into the sector.

Read the outlook

Industry perspective BCG · January 2026

The AI-First Biopharma Company

Move from deploy, to reshape, to invent.

The clearest sector-specific articulation of the progression from individual tools to redesigned workflows and genuinely different operating models.

Read the article

How to use this collection

A reading list, not a literature review

Sources are included because they illuminate a consequential question for Insights. The labels distinguish the maturity and provenance of the evidence; they are not a ranking of how interesting a source is.

Peer reviewed Foundational paper Preprint Working paper Research report Expert essay Industry perspective