A.I. Goes to War + Is ‘A.I. Brain Fry’ Real? + How Grammarly Stole Casey’s Identity

Episode Summary

Executive Summary: The episode examines three linked AI stories: the use of Claude and other AI systems in the Iran war, a study finding “AI brain fry” among workers using AI heavily, and Grammarly’s controversial Expert Review feature that impersonated real writers and editors. Across segments, the hosts argue AI is rapidly moving from productivity tool to military, psychological, and commercial power shift, often ahead of safeguards, consent, and trust.

Main Topics: AI in the Iran war (Priority: 5/5): The hosts discuss how U.S. and Israeli militaries are using AI for intelligence, target selection, logistics, and rapid battlefield analysis, with Claude integrated into military systems via Palantir’s Maven Smart System. Claude and the militarization of frontier models (Priority: 5/5): They focus on Anthropic’s Claude as a key model already embedded in classified military workflows, raising concerns about how consumer AI models are becoming part of warfighting infrastructure. Infrastructure as a wartime target (Priority: 4/5): The conversation shifts to Iran striking data centers and the fragility of AI/cloud infrastructure, including Middle East data center investments, fiber links, and supply-chain disruptions affecting semiconductors and cloud services. AI brain fry and workplace fatigue (Priority: 5/5): A segment with BCG researcher Julie Beddard explores survey findings that heavy AI use can create mental fatigue, information overload, and cognitive strain distinct from burnout. How organizations should adapt to AI (Priority: 4/5): The guests debate whether workers and managers should redesign workflows, use AI collaboratively, and set clearer outcome-based practices rather than piling on more tools. Grammarly’s Expert Review controversy (Priority: 5/5): Casey Newton describes being used without consent as a supposed expert in Grammarly’s AI writing feature, which generated generic advice under the names of real journalists and writers. The broader collapse of old SaaS moats (Priority: 4/5): The hosts argue that standalone products like Grammarly may be vulnerable as general-purpose chatbots and AI suites replace narrow paid utilities.

Key Arguments: AI in the Iran conflict is not hypothetical or future-facing; it is already being used operationally for intelligence processing, target suggestion, and faster battle planning. The military’s AI use is likely centered on “shrinking haystacks” of intercepted data rather than fully autonomous weapons, but the line between assistance and decision-making is blurring. Claude appears to be deeply embedded in military systems through Palantir’s Maven Smart System, making Anthropic’s model unusually central to classified defense operations. Military and tech companies have incentives to exaggerate capabilities, so reports about AI in war should be treated cautiously, though the hosts believe the operational role is real. AI can create a distinct kind of cognitive strain at work—“AI brain fry”—caused by oversight, multitasking, and pressure to keep up with AI-enabled expectations. Brain fry is different from burnout: it is more about cognitive overload and mental effort than emotional exhaustion, and it may be mitigated by better workflow design and team collaboration. Products like Grammarly are increasingly threatened because general-purpose AI tools can already do grammar and writing assistance more flexibly, cheaply, and with better models. The Expert Review feature was especially problematic because it attributed generic, hallucinatory advice to real people without consent or compensation. The hosts see a pattern in AI companies abandoning earlier anti-military or anti-surveillance principles once market incentives and defense contracts become attractive. Workplace and military AI adoption both raise the same governance question: how much decision-making should be delegated to systems that are improving faster than institutions can regulate them?

Data Points: AI quiz result: Slightly more people preferred AI-written passages than human-written ones - Discussed in the opening segment about the New York Times quiz comparing AI and human writing Brain fry prevalence among AI users: 14% - BCG survey result shared by researcher Julie Beddard Survey sample size: 1,488 workers - Workers surveyed in January for the BCG study Study duration: 8 months - UC Berkeley study referenced in Harvard Business Review on workplace intensity and AI Three-tool threshold: 3 to 4 AI tools - BCG finding that moving from three to four tools marked a sharp increase in strain Marketing disruption: 90% disrupted from a skill perspective - Julie Beddard citing prior work on which jobs changed most due to AI Grammarly subscription price: $144 per year - Casey Newton said users pay this for the feature set that included Expert Review Data center attack date: March 1 - The Guardian report cited by the hosts about Iran striking Amazon data centers in the UAE and Bahrain Targeted data centers: 2 in the UAE and 1 damaged in Bahrain - Iranian response to U.S. attacks, per reporting discussed on the show Potential civilian death toll: Over 175 people, mostly children - Iranian officials’ claim about a missile strike at an elementary school; hosts noted the incident was still under investigation Military integration date: 2024 - Claude integrated into Palantir’s Maven Smart System since that year Product lineup shift: 3 AI models - Hosts noted OpenAI, Gemini, and Claude as options now available to the Pentagon, expanding beyond Claude alone

Pivotal Quotes: "You're garbage, and I hope you lose your job and become homeless." — User messages cited by the hosts: Opening anecdote about backlash to the AI writing quiz "We are not there yet." — Kevin Roose: On autonomous weapons; he said current military AI use is still mostly intelligence, logistics, and target assistance rather than fully autonomous killing "AI brain fry is distinct." — Julie Beddard: She clarified that brain fry is a separate cognitive-fatigue phenomenon and is not the same as burnout

Implications: AI is already shaping war, work, and consumer software in ways that outpace consent and regulation. Expect more pressure on companies, managers, and militaries to justify how AI is used, who benefits, and who bears the risks.