Part I: Diagnosis

The Three Brain Networks Behind Every Career Decision

Before the specialty data, Castro explains the neuroscience. The brain operates through three large-scale networks that determine how physicians perceive and respond to disruption. The Default Mode Network (DMN) is the imagination engine—the 'what if?' network that runs differential diagnoses in the background. The Central Executive Network (CEN) is focused, goal-directed execution—the decade-long training target of medical education. But the pivotal one is the Salience Network (SN), anchored in the right anterior insula, whose job is to detect what matters and switch the brain between DMN and CEN. Research by Goulden et al. (2014) confirmed the SN switches intensify during harder decisions. AI, Castro argues, is the largest salience signal of a physician's career—and the physicians who are dismissing it have salience networks dulled by routine, not freed by rational analysis.

Specialty-by-Specialty: The Current State

Radiology is ground zero: 76% of all FDA-cleared AI devices (1,104 of 1,451) are in radiology (FDA, 2025). AI-flagged X-rays are read 20–30 minutes faster at Level I trauma centers. Viz.ai, deployed across 1,600+ hospitals, cuts stroke evaluation time by an average of 66 minutes—translating to an estimated 1.9 million neurons saved per minute of delay (Saver, 2006). Pathology's AI market will grow tenfold to $1.16 billion by 2035; ArteraAI's prostate tool, validated in Phase 3 RCTs with 11+ years of follow-up, now predicts 10-year cancer outcomes. Cardiology has 71 FDA-cleared tools; Caption AI enables nurses with minimal training to obtain diagnostic-quality echocardiograms. Mayo Clinic's dermatology program demonstrates that the human-AI team outperforms either alone—the single most important finding in all of medical AI research. Emergency medicine benefits from ambient documentation (reducing note-writing by hours per shift), AI triage, and real-time differential diagnosis support. Primary care AI adoption for research summarization has surged by 33 percentage points since 2023 (AMA, 2026).

The Antifragility Framework

Drawing on Nassim Nicholas Taleb's 2012 concept from Antifragile, Castro maps three physician career types. Fragile physicians double down on the exact competencies—fast pattern recognition, guideline memorization—that AI executes at scale, and will break when disruption arrives. Robust physicians tolerate AI tools without benefiting from them, staying relevant but not gaining strength. Antifragile physicians get better because of AI: they offload mechanical work and invest more deeply in what AI cannot do—empathy under uncertainty, ethical reasoning, complex human navigation. The goal of the book, Castro states plainly, is to move the reader from fragile to antifragile.

The Pattern Behind the Pattern: What AI Cannot Do

AI excels at pattern recognition from structured data, documentation, risk scoring, image interpretation, and workflow optimization. What AI cannot replicate: empathy under uncertainty (sitting with a patient after a terminal diagnosis), navigating family dynamics, breaking bad news with human presence, physical examination nuance, ethical judgment in ambiguous cases, and the clinical gestalt built from decades of embodied experience. Castro draws the defining distinction: your moat is not clinical knowledge (facts, guidelines, criteria—all areas where AI is superior) but clinical wisdom: applying knowledge to a specific human being's life, fears, values, and circumstances. The AI Impact Pyramid organizes work into three levels: replaceable tasks at the base (data entry, routine reads), augmented work in the middle (AI-assisted diagnosis and planning), and irreplaceable human judgment at the apex.

What's New — Q2 2026

1. 81% of Physicians Now Use AI Professionally — Double the 2023 Rate
The AMA's 2026 Physician Survey on Augmented Intelligence, drawn from 1,692 doctors across specialties and practice settings, found that 81% now use AI professionally — more than double the 38% reported in 2023. The top use case is summarizing medical research and standards of care (39% of respondents, up 33 percentage points from 2023), followed by creating care plans or progress notes (30%) and billing documentation (28%). Seven in ten physicians see AI as a tool to automate tasks that contribute to burnout, while 76% say it enhances their ability to care for patients.

2. FDA Reaches 1,451 Cleared AI/ML Medical Devices — 295 in 2025 Alone
The FDA cleared a record 295 AI/ML-enabled medical devices in 2025, up from 253 in 2024 and just 6 in 2015, bringing the cumulative total to 1,451 devices since tracking began in 1995. Radiology continues to dominate, accounting for 76% of all cleared devices (1,104 cumulative), while cardiovascular and neurology applications are expanding. In Q4 2025 alone, 72 devices were cleared — 55 of which were radiology tools — reflecting steady acceleration in AI's clinical footprint.

3. Randomized Trial: Both AI Scribes Reduced Burnout, One Outperformed on Time Savings
A randomized crossover trial published in JAMIA (February 2026) involving 160 outpatient clinicians at a tertiary academic medical center found that both tested ambient AI scribe products significantly reduced personal and work burnout scores on the Copenhagen Burnout Index and improved workflow satisfaction. Product B showed superior performance: it reduced average minutes-in-notes per day by 3.19 more minutes than Product A (95% CI: −4.87 to −1.50). The study is one of the first head-to-head comparisons of competing ambient scribe technologies at scale.

4. Ambient Scribe Cuts Documentation Time 15%, Boosts Eye Contact — Even in Asia
A prospective study at Singapore General Hospital published in JMIR Medical Informatics (March 2026) found that experienced ambient scribe users achieved a 15.0% reduction in documentation time per consultation (from 5.3 to 4.5 minutes; P=.04) and a 10.6% increase in eye contact time. Importantly, consultation duration was unchanged — clinicians redirected effort toward patient engagement rather than speeding through visits. Of 39 surveyed patients, 69.2% agreed their physician focused on them more, and none expressed discomfort with the technology.

5. AMA Survey: 88% of Physicians Worried About AI-Driven Skill Atrophy
While enthusiasm for AI is growing, the 2026 AMA survey reveals a counterweight concern: 88% of responding physicians reported at least some worry about AI-related skill loss, with 70% "very" or "somewhat" concerned specifically about skill erosion in medical students and residents being trained today. Additionally, 85% of physicians want to be consulted or directly involved in decisions about AI adoption at their institutions — suggesting clinicians want to steer, not just ride, the AI wave.

Sources: American Medical Association 2026 Physician AI Survey, The Imaging Wire — FDA AI Device List Q4 2025, IntuitionLabs FDA AI/ML Device Tracker, JAMIA — Randomized Crossover Trial of Ambient Scribes (Feb 2026), JMIR Medical Informatics — Ambient AI Scribe Singapore Study (Mar 2026)