Henry Faurd

Carta's Plugins for Claude are live. All three. 🎉 The two biggest things I've been focused on for the past couple of years: the ERP and AI. This is what happens when they come together. Investors, Cap Table, and CRM, generally available to every Carta customer today. Fund data lives in one system, cap tables in another, and LP relationships in a third. Carta's ERP already connects all of it. Now Claude can work across all of it too, in one conversation: Open Claude, pull performance against benchmarks, check the cap table, see what your IR team already discussed with your LPs, and ask it to draft the update. One session. Live data. Three days of data assembly, gone. Excited for our customers to try it out and please send feedback! https://lnkd.in/gwcvfZbW

Last week, I invited a group of senior engineers shaping AI direction up to Tahoe for 3 days with only one instruction: build something ambitious with AI you wouldn't normally have time for. One team built AI agents to run meaningful parts of fund administration autonomously, using Carta’s internal tools and CLI to investigate a case, decide what to do, and take actions such as booking a journal entry, record its reasoning, and prepare the customer response. A second team built a software factory that can autonomously take on engineering work ranging from production bugs to larger code changes, spin up coding agents with the tools and application environments needed to implement and verify the work, and open pull requests. In one run, it migrated an entire application page to a newer frontend framework. The team has already reduced agent startup time from about 20 minutes to five or six. A third group explored whether smaller, more tailored models could be better suited to some of our workflows. They built the infrastructure to train, serve, and evaluate them. The first models did not outperform the frontier models we use today, but we now have a repeatable way to test where this approach may make sense. Some of this will ship soon. Some of it may not. That wasn't the point. The point was giving three good teams 3 days and room to experiment.

Before my first day with Friends for Youth (FFY), the trainer gave me one piece of advice: most adults talk too much. Just listen. The FFY program pairs adult mentors with kids who could use another consistent person in their life. I joined 13 years ago intending to get practice being a father before my son was born. They matched me with a 13-year-old named Joevanny. Our first outing was an hour drive to see the sea lions at Año Nuevo. Two strangers in a car, and a lot of awkward silence. On the way home, he told me he hated school. I didn't say the things adults usually say ("stick with it, school’s important"). I just asked why, and let him talk the rest of the drive. That became our pattern for a year. He'd bring something up — trouble at school, an unfair teacher, something at home — and I'd resist the urge to fix it or lecture him. I'd just ask a question and listen to the answer. I've been a father for 12 years now, and a CEO longer than that. Both jobs have convinced me the trainer's advice applied to more than just connecting with teenagers. Most advice-giving is really just impatience. The people who listen longer usually understand more, and understand it faster, than the people who jumped in first.

Our frontend team migrated over 200 Carta applications from React 17 to React 18. We projected six months. But it only took one week. React is the technology that runs every screen in our product. Version 17 is officially End of Life, which introduces security risks, misses out on modern performance gains, and blocks our teams from tapping into the modern frontend ecosystem. Carlton Howell and Denis Kumbaradzi on our frontend platform team built and executed an entire React upgrade playbook end-to-end over a single hackathon. Their playbook: build out a multi-step execution plan and a tailored Claude skill before writing a single migration PR. This foundation relied on three things: our past internal upgrades, open-source community learnings, and the team's own read on what would actually work. They started with our 40 shared frontend libraries, then rolled straight through all 200+ apps, beginning with our largest flagship platforms. Claude orchestrated the migration through concurrent agent workers via worktrees, driving parallel batches of complex code transformations simultaneously. The part that compounded: after every batch, Claude reflected on what it learned (a testing quirk, an edge case, something worth scripting) and fed it directly into the next run, autonomously opening PRs to refine its own skill file as it progressed. Batch 40 ran faster than batch 4. The result was 86 multi repo PRs and 56,000 lines of non-trivial code. Average merge time: Just over 1 hour. I've watched the video the team made about this more times than I'll admit. It's amazing. 😂 Watch it below. React 19 is next.

Startups are raising 50% more capital than in 2023. Headcount is flat. At the early stage, companies are half the size for the same dollars raised. We are living through a complete rewrite of how companies are built and scaled.