Skip to content

Research and essays

Building the AI Scientist

An eight-post arc developing the category from first principles: what an AI scientist is, the cognitive architecture it requires, how to build it from validated work loops, why model plus data is insufficient, and what changes in the operating model of scientific work.
Eight small paper forms in a row progressing from a loose pile of scraps on the left to one crisply folded resolved form on the right.
The arc: eight posts from scattered evidence to a resolved position. Two published, six forthcoming.

Published · 2 of 2

ESSAY 0121 July 2026 · 9 min read

Building the AI Scientist for the Hardest Part of Science

A true AI scientist must connect two fundamentally different capabilities: constrained computation and causal reasoning across incomplete scientific knowledge.

DefinitionScientific reasoning

Top-level claim

An AI scientist must master both map problems with clearer destinations and detective problems built from incomplete clues.

Audience

Heads of R&D, CIOs, CDOs, AI strategy leaders

Read the essay
ESSAY 0222 July 2026 · 12 min read

The Cognitive Architecture of an AI Scientist: Inside the Scientific Harness

An effective AI scientist depends on cognitive systems that compute, interpret, remember, evaluate, and work with people as one calibrated architecture.

ArchitectureEvaluationHuman collaborationDemonstrations

Top-level claim

An AI scientist becomes effective when runtime and control, scientific cognition, scientific judgment, and reviewable work products operate as one calibrated architecture.

Audience

R&D leaders, AI product leaders, architects, engineers, computational scientists

Read the essay

Forthcoming · 6

  1. ESSAY 03

    Build From the Work Loop Up, Not the Model Down

    AI scientists should be built from the vertical work loop upward, not the generic model downward.

    Work loops
  2. ESSAY 04

    Why Model + Data Is Not Enough for Scientific Work

    A powerful model plus proprietary data can answer questions, but it does not by itself know how to do scientific work.

    Enterprise deploymentArchitecture
  3. ESSAY 05

    Beyond Research: The AI Analyst Across Pharma

    The AI scientist is the research-facing expression of a broader AI analyst spanning three forms of science-backed knowledge work.

    Enterprise deployment
  4. ESSAY 06

    What Can the AI Analyst Automate?

    The degree of automation depends on ambiguity, novelty, stakes, reversibility, evaluation and accountable human judgment.

    Human collaborationEvaluation
  5. ESSAY 07

    Who Has the Right to Build the AI Analyst?

    The right to build must be earned through deep science, deep AI systems expertise, and proof in real work.

    Enterprise deployment
  6. ESSAY 08

    The AI Analyst Changes the Operating Model of Science-Backed Work

    The AI analyst changes who performs, reviews, decides, corrects, archives and learns from science-backed work.

    Human collaborationEnterprise deployment