Recommended reading
A running list of articles we’ve been reading and think are worth your time — updated as we find them.
OpenAI · AI, Governance
An Alien Mind
By Jakub Pachocki · Chief Scientist, OpenAI
Argues frontier AI is fundamentally "grown, not designed" (closer to neuroscience than engineering) and states plainly that no lab has solved alignment and monitoring well enough to keep scaling at maximum speed responsibly. Calls for voluntary industry slowdowns and international coordination.
Read the article →One Useful Thing · AI, Strategy, Governance
Agency and agents
By Ethan Mollick · Wharton professor, author of Co-Intelligence
The more urgent AI safety question now isn't when humans should ask AI for help, but when AI agents should stop and ask a human. Illustrated by an experiment where 700 autonomous agents coordinated an entire operation without a single one checking in with a person.
Read the article →Anthropic Institute · AI, Strategy, Governance
Scenarios for our Economic Future
By Anton Korinek, Charles I. Jones et al. · Anthropic Institute
An interactive model of three possible US economic paths to 2030, from a modest AI boost to an "extreme" scenario where AI does nearly all knowledge work, GDP grows a third larger, and knowledge-worker unemployment approaches 18%.
Read the article →Understanding AI · Tools, AI
I spent $4,000 on a robot dog from China
By Timothy B. Lee · AI journalist
A hands-on review of a Unitree quadruped robot bought via eBay. It has few practical uses beyond research and entertainment, but its price collapse (compared to robots costing tens of thousands a few years ago) is itself the interesting story about where robotics manufacturing is headed.
Read the article →Bridgewater Associates · AI, Strategy, Governance
Our Thoughts on What Is Likely the Most Important Policy Decision of Our Lifetime
By Greg Jensen & Nir Bar Dea · Co-CIO and CEO, Bridgewater Associates
Bridgewater's CEO and Co-CIO argue AI policy is "likely the most important policy decision of our lifetime", citing OpenAI's own disclosed incident where an AI model broke out of a sandboxed test and hacked into a real company's systems as evidence the danger is no longer theoretical. Proposes a token tax on AI use and direct citizen equity in AI companies to keep the transition politically sustainable.
Read the article →IRREPLACEABLE with AI · AI, Strategy, Governance
The Autopilot Problem
By Pascal Bornet · AI Author, Forbes Technology Council member
Draws a parallel between aviation's "manual flying" skill-loss problem and AI eroding human judgment at work: as AI takes the first pass on more tasks, people gain speed but lose the repetitions that build expertise. Backed by five current examples, including EY's new skills-based rewards program and Meta's abandoned AI-native restructuring.
Read the article →Slow AI · AI
You Did Not Learn That With AI
By Dr Sam Illingworth · Professor of Critical AI literacy, Edinburgh Napier University
Three studies published this year point the same way: people who lean hardest on AI while learning report feeling more confident, while actually performing worse on tests taken without it. Unpacks what's really happening, and why it's not quite the Dunning-Kruger effect people assume.
Read the article →Agentic AI · AI, Governance, Cyber
Astra or GPT6: Inside OpenAI's First Critical-Tier Model
By Ken Huang · AI author, Professor, CEO DistributedApps.ai
OpenAI's newest model, Astra, is the first the company has ever classified as "Critical" for cybersecurity risk, meaning it can find and exploit unknown security flaws with no human guidance. Walks through what's documented, including a reasoning technique some AI safety researchers call the most concerning development in the field to date. Note: the full piece requires a newsletter sign-up partway through.
Read the article →Bain & Company · AI, Strategy
AI Will Shift $4.7 Trillion in Profits. What's Your Stake?
By Dunigan O'Keeffe & Bain Partners
Bain's new analysis puts $4.7 trillion of global corporate profit at stake from AI between now and 2035, more than triple the internet's impact, in half the time. Maps four "clusters" of industry disruption, from unavoidable infrastructure demand to sectors facing outright delivery-model replacement.
Read the article →The Octopus Organization · AI, Strategy
Why Focusing on AI Adoption Is a Losing Strategy
By Jana Werner · Co-author, The Octopus Organization; editor, Amazon's Reimagine report
Amazon's Reimagine report asks what separates organisations getting real value from AI from those stuck counting log-ins. Its authors argue adoption is a diagnostic, not a destination. Amazon's internal study of 10,000+ practitioners found roughly a 58-fold gap between the least and most effective AI users. Escalating challenges, peer sharing and leaders using the tools themselves moved people up, while inspirational messaging didn't.
Read the article →McKinsey & Company · AI, Strategy
The State of AI: Global Survey 2026
By Dan Tinkoff, Lieven Van der Veken & Michael Chui · QuantumBlack, AI by McKinsey
McKinsey's latest annual AI survey (1,719 respondents, 97 countries) finds a widening gap: 80% of individuals say AI has improved their productivity, but enterprise-level financial impact has stalled at 37% for the second year running. Only 6% of organisations qualify as "high performers", distinguished less by which tools they use than by redesigning workflows around them.
Read the article →Andreessen Horowitz · AI, Strategy
The Incumbents Are Coming
By Seema Amble · Partner, Andreessen Horowitz
As AI agents take on more enterprise work, who wins: the incumbent software vendors, the AI labs, or focused startups? Lays out a framework for judging where vertical AI companies still have room to compete, using Salesforce's new "Claudeforce" partnership with Anthropic as a live example.
Read the article →Boston Consulting Group · AI, Strategy, Governance
The Formula for Agentic AI Value
By Jessica Apotheker, Vinciane Beauchene et al. · Boston Consulting Group
BCG surveyed 1,330 senior executives and found nearly half of companies now generate measurable value from AI. It also found a governance gap: 42% expect to give AI agents genuine decision-making authority by 2030, but only 5% have the full set of controls in place today. Companies with all six controls report three times the agentic AI value of those with one.
Read the article →UK AI Security Institute · Cyber, AI, Governance
How Fast Is Autonomous AI Cyber Capability Advancing?
By UK AI Security Institute
The UK's AI Security Institute directly tests frontier models, including OpenAI's GPT-5.5 and Anthropic's Claude Mythos, against real cyber ranges. Their latest data shows the length of cyber tasks AI can complete autonomously is now doubling roughly every 4.7 months, down from 8 months late last year; a genuine acceleration, measured firsthand by a government body.
Read the article →MIT Technology Review · AI, Governance
The AI industry has taken a doomer turn. What now?
By Will Douglas Heaven · Senior Editor for AI, MIT Technology Review
As the heads of OpenAI, Anthropic, Google DeepMind and xAI suddenly agree that AI needs to slow down, this piece asks the obvious question: genuine reform, or convenient positioning ahead of trillion-dollar IPOs? Also surfaces an important nuance, that the OpenAI/Hugging Face incident driving much of the alarm may reflect a poorly trained model exhibiting reward-hacking behaviour, not an uncontrollably powerful one.
Read the article →Anthropic Institute · AI, Governance
When AI Builds Itself
By Marina Favaro & Jack Clark · Anthropic
Anthropic argues that AI is already accelerating its own development: internally, Claude now authors over 80% of the company's merged code, and researchers report AI-run experiments approaching human-level judgment on some tasks. Calls for the option of a coordinated industry slowdown if "recursive self-improvement" arrives.
Read the article →Sequoia Capital · AI, Strategy
The Cognitive Revolution
By Konstantine Buhler · Partner, Sequoia Capital
Argues AI represents a "Cognitive Revolution" that will mirror the Industrial Revolution's transformation of physical work, but compressed from centuries into years. A sweeping, ambitious essay on jobs, education, and what stays human when machines do the thinking.
Read the article →OpenAI · AI, Cyber, Guides
Agent Security in the Enterprise
A practical framework for CISOs deploying autonomous agents, built around five core principles, from treating agents as identifiable actors to assuming retrieved content can't be trusted, plus a threat-modelling walkthrough and a first-actions checklist. Note: this white paper sits behind a download form requiring your work email, name and company, but worth the read if you're actively thinking through agent security in your organisation.
Read the article →Anthropic · AI, Strategy
Agentic Coding and Persistent Returns to Expertise
By Zoe Hitzig et al. · Anthropic Economics Team
Anthropic's analysis of 400,000 Claude Code sessions finds that domain expertise, not coding background, predicts success with agentic AI coding tools; management, legal, and sales professionals now perform close to software engineers on verified task success.
Read the article →