Writing and conversations published elsewhere - on AI risk in legal practice, data governance, and the decisions lawyers are navigating right now.
Venture capital has quietly shifted from financing innovation to financing distribution. The result: legal AI companies built on rented APIs, thin interfaces, and speculative pricing - valued as if they've solved what they've only wrapped.
AI-first isn't a strategy - it's a symptom of confusion. Firms that deploy AI before fixing their workflows, data discipline, and incentive structures don't get transformation; they get an expensive mirror showing exactly what was already broken.
Six fictional but technically grounded scenarios showing how AI systems used in legal work can be exploited — through prompt injection, goal hijacking, attention bias, adversarial audio, and poisoned knowledge bases. Understanding the attack surface is the first step to defending it.
A ground-level account of where the legal profession actually stood with AI in 2025 - not the hype, but the pattern: edge adoption, unresolved governance, and the growing realization that organizational readiness, not model capability, is what limits progress.
The legal AI conversation in 2026 feels settled - and that's the problem. A clear eyed look at where confidence is outpacing evidence: reasoning plateaus no one wants to name, AGI timelines that don't match how labs actually build, labor disruption that isn't materializing, and a security threat profile expanding faster than most firms realize.
The building blocks for AI-era innovation are already inside most law firms. A walk through planning, deploying, and driving adoption of technology in a business of law — incremental steps, incremental gains, and experimentation anchored in strategic positioning.
When AI is part of how legal work gets produced, firms need to be able to show who authorised it, how outputs were reviewed, and where responsibility sits. Most cannot.
The distinction between chatbots and AI agents, what security vulnerabilities mean in practice for law firms, and why asking non-technical users to detect prompt injections is not a reasonable security posture.