# Bruce Booth says AI makes pharma R&D worse first. a16z disagrees. I side with Booth.

> Bruce Booth expects AI to hurt pharma R&D productivity for three to five years. a16z disagrees. Why crowded targets favor Booth.

URL: https://www.ch-healthtech.com/insights/booth-vs-a16z-ai-pharma-rd-productivity-worse-first
Markdown: https://www.ch-healthtech.com/insights/booth-vs-a16z-ai-pharma-rd-productivity-worse-first.md
Published: 2026-11-19
Author: Christian Hein
Tags: technology/artificial-intelligence, industry/large-pharma, industry/tech-bio, industry/venture-capital, function/clinical-development, function/pre-clinical-research

---


## TL;DR

Bruce Booth predicts AI will make pharma's R&D productivity worse for the next three to five years, and Andreessen Horowitz argued the opposite last week. I think Booth is right about the industry for the next few years. I also think a16z will be right about a handful of companies. Today's models design well around familiar targets, so more candidates mean more companies competing for the same capacity, trial sites and patients.

AI will make pharma's R&D productivity worse for the next three to five years. That is Bruce Booth's prediction at Atlas Venture. Andreessen Horowitz argued the opposite last week, and I think Booth is right.

Both expect AI to produce far more drug candidates. For a16z, that means more shots on goal. The authors also think better models can improve the odds that each candidate succeeds and shorten the time to approval. If they're right, the cost per approved drug comes down.

## Where the extra candidates would go

Booth asks where all those extra candidates are going. Today's models are particularly good at designing molecules around familiar targets, because that's where the training data is richest. The industry already has more than 100 programs against each of the top eight targets worldwide. At least 30 targets have more than 30 programs each. Add more candidates to those pipelines and more companies compete for the same toxicology slots, CMC capacity, trial sites and patients.

The a16z authors are betting on "novel drugs that simply wouldn't have existed before". Booth looks at AI-native pipelines today and points to familiar targets including TSLP, HER2, KRAS and TYK2. A new drug against a familiar target can still be a better medicine. It may take years of clinical trials to find out which one is better, though, and every company in the race has to pay for its own trials.

Daphne Koller puts some numbers on the development side. She estimates that AI could cut the time from candidate nomination to launch by 11–17%. Her model assumes faster IND-enabling work and trial operations. About half the timeline is spent dosing and following patients, and she holds that time fixed. She also sees a way to shorten it. A better biomarker could tell us months earlier whether a drug is changing the course of a disease.

## Whose productivity is being measured

Part of the gap between Booth and a16z comes down to whose productivity you're measuring. An AI-native company could pick a better candidate, get it to patients faster and earn an exceptional return. The industry still bears the cost of everyone else who pursued the same target. Both outcomes can happen at once.

On Tuesday I wrote that Isomorphic Labs could change pharma completely if it solves the data problem in human biology, and a16z has invested in Isomorphic's $5 billion Series C. Isomorphic says its first programs are heading toward clinical trials. It has disclosed preclinical work in oncology and immunology, but we still know little about the individual candidates. Maybe the investors know something we don't.

I think Booth is right about the industry for the next few years. I also think a16z will be right about a handful of companies. Isomorphic could be one of them.

For me, the hardest problems are still finding genuinely new targets that matter in human disease, improving the odds that a drug works in patients, and finding out sooner when it doesn't.

