Detailed Analysis
Andrew Wilkinson, a serial entrepreneur and prolific business acquirer, describes a dramatic shift in his operational approach beginning in December 2025, when he became deeply immersed in Anthropic's Claude Code — referred to throughout the podcast transcript as "OpenClaw" and "Cloud Code," artifacts of imprecise speech-to-text transcription. Wilkinson characterizes the experience in visceral terms, describing himself waking at 3 or 4 in the morning to work in terminal sessions with Claude Code, a behavioral change he frames as both compulsive and transformative. His central claim is that he has been running a SaaS business almost entirely autonomously using Claude-powered agents, using the experiment as a proof of concept before deploying similar automation across his broader portfolio of operating companies.
The practical applications Wilkinson describes span both personal and professional domains. On the business side, he identifies administrative burdens — customer support, accounting, routine written communications — as the primary targets for AI-driven automation, arguing that these repetitive, rules-based tasks are now ripe for displacement. A telling anecdote involves a trip to Arizona during which he forgot his laptop entirely but was able to manage his business operations through a Claude agent accessed from the back of ride-share vehicles, with no external party apparently detecting that his correspondence was AI-generated. On the personal side, he describes building a psychological assessment tool using Claude Code that synthesized personality test data for himself and his partner, eventually spinning the concept into a commercial product called Deep Personality. The ease and speed of that build — roughly 40 minutes to produce a scored, JSON-formatted test suite — illustrates the low barrier to prototyping that current AI coding tools have created for non-engineers.
Wilkinson's candid accounting of his productivity breakdown is among the more analytically honest assessments of agentic AI workflows to appear in mainstream business media. He estimates that approximately 50 percent of his time is consumed by debugging, 30 percent by iteration and improvement, and only 20 percent by genuinely productive output — a ratio he himself labels the "classic productivity treadmill." This admission complicates the utopian framing common in AI adoption narratives and points to a maturity gap between the demonstrable capability of tools like Claude Code and the friction costs of maintaining and extending agentic systems at scale. The metaphor he reaches for — "chasing the dragon," referencing an initial peak experience that proves difficult to replicate consistently — captures a widely reported phenomenon among early adopters of autonomous AI workflows.
The episode situates itself within a broader and accelerating trend of high-net-worth operators and founders treating AI agents not merely as productivity supplements but as organizational infrastructure. Wilkinson's family office context is particularly significant: the deployment of AI agents in investment research, deal evaluation, and portfolio administration represents a frontier where the stakes and the complexity of automation are substantially higher than in consumer applications. His framing of AI as a tool for identifying "next generation" business categories and capital allocation decisions suggests that the competitive differentiation in venture and private equity may increasingly hinge on the sophistication of an investor's AI operational stack. The episode, though informal in register, functions as an early-adopter case study in what autonomous business management might look like as Claude and comparable systems continue to improve.
Read original article →