Mark Williams·May 22When Optimization Optimizes ItselfRecursive Goodhart’s Law in Self-Modifying AI Systems
Mark Williams·May 15Context is Infrastructure, Not InstructionsWhat teams gain when they govern AI context like a software dependency
Mark Williams·May 8The Turn as the Unit of QualityWhat makes iterative refinement productive, and when it starts to hurt
Mark Williams·Apr 30The Capability-Reliability Split in Agent SystemsWhy frontier agents reach state-of-the-art on one run, and fail at the same task on the next
Mark Williams·Apr 23The Edge of the UnderdefinedThis is the final article in “The Meta-Engineer,” a three-part series examining how AI is reshaping the identity and skill set of software…
Mark Williams·Apr 16Don’t Vibe, ArchitectHow professionals work with agents, how context scales, and why orchestration is a transitional skill
Mark Williams·Apr 9Context is the New CodeThis is the first article in “The Meta-Engineer,” a three-part series examining how AI is reshaping the identity and skill set of software…
Mark Williams·Apr 1Closing the Loop: How Human Corrections Can Make AI Systems Smarter Over TimeEvery day, thousands of domain experts in law firms, hospitals, and financial institutions review the outputs of AI systems and quietly fix…
Mark Williams·Mar 25Offline RL and the Data FlywheelHow production systems learn from logged data, and why dataset quality is the most underinvested layer of the RL stack
Mark Williams·Mar 18Reward Design as ArchitectureWhy the reward function is the most consequential, and most overlooked, design decision in any RL system
