A VC Built an AI Chief of Staff That Tracks 100 LP Contacts and Improves Itself Weekly

Ryan Sarver, a venture capitalist in the middle of a fundraise, published a post last week detailing how he built an AI chief of staff that, in his telling, outperforms every human he has ever hired for the role. The post drew 757,000 views on X. Garry Tan, president of Y Combinator, followed with a public endorsement: “I’m telling most everyone I know that they should build a personal OpenClaw,” according to Yahoo Tech.

The system Sarver built runs on Claude Code, Anthropic’s agentic AI environment. It is assembled from flat markdown files, Python scripts, and an LLM layer handling synthesis and judgment. It tracks more than 100 LP contacts across a fundraise, preps Sarver before every meeting via WhatsApp, extracts action items from meeting notes, and maintains a relationship CRM that updates continuously. The most distinctive component is what Sarver calls “Kaizen”: a weekly self-improvement loop where the system scans what other builders in the Claude Code community are doing and surfaces patterns worth adopting. Per Yahoo Tech, “the whole thing gets measurably better every week without him actively tinkering.”

Who Benefits First

Yahoo Tech and Forbes framed Sarver’s post as evidence that a new class of knowledge worker is emerging. The early adopters are not generalists looking to write emails faster. They are people who already think in processes: VCs managing deal pipelines, operators running portfolio companies, analysts synthesizing large data sets.

Boris Cherny, the creator of Claude Code at Anthropic, predicted in February that the tools would expand into “pretty much any kind of work that you can do on a computer.” He described the professionals who would thrive as people who are “curious and generalists, and they cross over multiple disciplines and can think about the broader problem they’re solving rather than just the engineering part of it,” according to AfroTech.

The Career Ladder Problem

The labor market implications cut both ways. A February 2026 analysis from the Dallas Fed found that returns on job experience are increasing in AI-exposed occupations because AI can replicate codified knowledge but not the tacit knowledge that experienced workers carry. What AI automates most cleanly is the work that entry-level and mid-level knowledge workers do: research synthesis, meeting prep, relationship tracking, document production.

That is exactly the work Sarver’s system now handles. It is also the work that historically served as the training ground for developing judgment. Entry-level analysts at consulting firms learn to think by doing research. Junior VCs learn to evaluate companies by sitting in on calls. As the Dallas Fed analysts noted, “firms are going to find that AI is making this method of employee development cost-ineffective, at least in the short run,” per Yahoo Tech.

Workers with advanced AI skills already earn 56 percent more than peers in the same roles without those skills, according to PwC analysis cited in Yahoo Tech’s reporting. The gap between operators who build agent infrastructure and those who do not is compounding weekly, in a way that spreadsheets and previous software tools never could.