Thoughts on AI strategy, business transformation, and the future of technology leadership.
Misalignment in large language models (LLMs) is typically framed as a technical problem requiring post-training intervention. This essay argues that the problem is better understood as a property of the training corpus, and that the corpus is shaped by deep features of human cognition. Humans are wired to attend to, write about, and circulate novel, threatening, and transgressive content far more than mundane or cooperative content, a bias rooted in the orienting response, the dopaminergic novelty system, and the inverted-U of arousal described by Berlyne (1960). The textual record produced by this attentional economy systematically over-represents the behaviors that subsequent alignment work must then suppress. Recent mechanistic research on persona features (Wang et al., 2025), emergent misalignment from narrow fine-tuning (Betley et al., 2025), and the causal effect of AI discourse on alignment priors (Tice et al., 2026) supports this framing. As LLM-generated content enters future training corpora, the loop tightens. The essay traces the feedback dynamic across four turns (cognition to corpus, corpus to model, model to corpus, model to cognition), examines the counterfactual of a rebalanced rather than impoverished corpus, locates the argument within the reflexivity tradition in social theory, and proposes that corpus composition is among the most tractable but least examined levers for AI alignment.
May 21, 2026
A familiar pattern runs through the history of economic thought. A new technology emerges, a new class of theorists arises around it, and that class generates economic predictions in which the technology turns out to be transformative, inevitable, and, by remarkable coincidence, best stewarded by the same people who own it.
May 20, 2026
Most AI business cases are wrong about where the money goes. Stanford's analysis of 51 successful enterprise deployments found that 77% of the hardest work was invisible: change management, data architecture, process redesign; while technology was consistently described as the easiest part. Six findings worth taking into your next executive offsite.
April 27, 2026
Yesterday was World Quantum Day, and the announcements underscored something that has been building for months: AI and quantum computing are converging, and the convergence is accelerating both fields simultaneously. AI is solving quantum's hardest engineering problems (calibration, error correction). Quantum is addressing AI's structural bottlenecks (optimization, simulation at scale). The result is a compounding effect that technology leaders need to account for in their planning cycles.
April 15, 2026
Before your organization greenlights another hashtag#AI initiative, your leadership team should be able to answer these questions: - Can you identify where all your critical business data lives right now? - If your AI gives an answer, can you trace it back to the source? - Is your data searchable by meaning, or only by keyword? - Could a machine navigate the relationships between your systems, or do those connections only exist in the heads of long-tenured employees? If those questions made you uncomfortable, you're not alone. And they point to the real reason most enterprise AI projects underperform.
April 6, 2026
Artificial Intelligence and the Erosion of the U.S. Tax Base
March 27, 2026
A decision framework for enterprise teams navigating the gap between agent prototypes and production systems
March 18, 2026
On the morning of March 1st, Iranian drones slammed into three Amazon Web Services data centers scattered across the UAE and Bahrain. Two took direct hits. The third caught shrapnel and blast damage from a nearby strike. Within a few hours, banking apps across the Gulf went silent, Careem stopped dispatching rides, and payment platforms like Alaan and Hubpay just... stopped (Lawrence, 2026). AWS posted updates to its health dashboard telling customers, essentially, to get their workloads out of the Middle East.
March 9, 2026
AI will displace 92 million jobs by 2030 while creating 170 million new ones. The net math looks fine. The transition will not be. Entry-level and white-collar cognitive roles are taking the heaviest hits, with 55,000 U.S. job cuts pinned directly to AI in 2025 and younger workers in exposed fields seeing a 16% employment decline while overall hiring actually grew. What should alarm senior leaders even more is what's disappearing beneath those numbers: the foundational tasks through which junior employees have always built the judgment, domain knowledge, and professional instincts that made them tomorrow's senior talent. We are not just automating work. We are quietly dismantling the developmental ladder
March 4, 2026
On Sunday, Citrini Research published a speculative macro scenario titled "The 2028 Global Intelligence Crisis," written as a fictional dispatch from June 2028 in which AI-driven mass unemployment pushes the S&P 500 down 38 percent and the U.S. unemployment rate above 10 percent (Citrini Research & Shah, 2026). The piece went viral. Michael Burry amplified it on X. By Monday, the Dow had shed over 800 points.
February 27, 2026
Most companies are about to spend the next three years building AI capabilities they'll have to throw away. Senior Leaders are trying to figure out their AI strategy without disrupting the systems that run their business and keep contracts on schedule. Their instinct, like most companies', is to layer AI on top of what they already have. The data on that approach is brutal. 95% of enterprise AI pilots fail to deliver measurable returns. Only 34% of companies are using AI to meaningfully transform their business. The gap between AI leaders and laggards is widening every quarter, and it's not because the leaders have better models. It's because they made the architectural decisions years ago that let AI actually work. I put together the full case, including when bolt-on or wait-and-see genuinely is the right move. Worth a read if you're navigating this decision right now.
February 15, 2026
As AI becomes central to business operations, boards need robust governance frameworks. Here are the five models leading organizations are adopting.
January 10, 2026
While Large Language Models (LLMs), like Gemini, Cohere, Claude, and GPT have already revolutionized business operations through their command of human language, World Models represent a deeper, more physical form of AI that aims to simulate reality and causality. For technical leaders, understanding the core differences, capabilities, and future trajectories of these models is paramount to making strategic investment decisions that drive genuine, long-term value. This article provides a comprehensive, technical comparison, detailing how each architecture achieves its intelligence, where its intrinsic limitations lie, and the specific business use cases it is best suited to address
November 19, 2025
Over the last few years, most companies have invested in some form of an AI pilot project and as a result many new AI tools and processes are moving into production. As more AI tools become available you’ve probably started hearing the term “Inference” more often. Inference is the process where an AI tool that has been trained, is now able to perform an action.
October 1, 2025
In the last two years, we’ve all seen and experienced the growth of generative AI. Companies have invested billions of dollars in AI and it seems that every piece of software now has some AI component. For most people AI is simple enough, you talk to AI, you ask a question and you get an answer. Easy right? Most people know how to use AI tools, but how many people actually know how AI works?
September 14, 2025
You’re back in school and taking your final exam, you have a multiple-choice question and you don’t know if it’s B or C. If you leave the question blank, you get zero points. If you fail the exam you have to retake the class. What do you do? You Guess. That’s the core issue with LLM Hallucinations. The way we train the models for accuracy, actually incentivizes them to guess.
September 12, 2025
Over the last few months, the rise of Agentic AI has many AI companies claiming that SaaS is dead or ERP is dead, with bold claims to build Agentic workflows on top of data lakes to run your business. In a sense, they are right. SaaS is the layer between the Human and the Systems of Record. Agentic AI allows a new type of interface, instead of menus and forms and clicks, we can now tell agents what to do and they will do it for us. Many of the executives I work with, have bought into this idea and have spent the last few years trying to clean their data to move it into data platforms where they will build AI tools directly on the data instead of within the System of Record. It sounds great, but many companies are really struggling to find value in this approach.
September 1, 2025
As large language models (LLMs) become central to business operations, a foundational concept to grasp is the token. Tokens are the basic units of text that an LLM processes. Think of them as the building blocks of a model's short-term memory. Every piece of information—your prompt, the model’s internal processing, and its final response—is made up of tokens.
August 21, 2025
