In the previous post, we presented the levels of VC AI adoption: the basic user (relying on consumer apps), the tech-savvy intermediate (writing local scripts and MCP connections), and the power user (building custom in-house solutions).
We also looked into the problems of doing everything in-house: eventually, when you build your own AI infrastructure, you quickly accumulate tech debt and shift your focus to maintenance instead of investing.
To break this ceiling, you should first understand what it means to run an AI-native firm.
What it actually means to be AI-native
Being AI-native means you stop treating AI as a tool you prompt for productivity or automations, and start treating it as infrastructure: a system with all relevant data available to it, continuously mapping and digesting the current state of everything.
Firms that build on these three principles save around 10 hours a week per employee, in our experience.
A unified knowledge layer
Your data across CRM, Drive, notes, and email gets digested into one queryable layer, so a question about your firm's history gets a sourced answer instead of a 30-minute dig through five tools. And your firm's "cookbooks" and "how-tos" are also shared across everyone, so a repetitive task never diverges from your firm's standards.
Tools like LlamaIndex, Unstructured, or even cloud services like Vertex AI RAG Engine can help you digest data into agent-friendly, queryable context. Note that you still need to manage the digestion pipeline, credentials, and permissions when doing it yourself. Claude for Teams allows you to share skills, which can help with repetitive tasks, but those skills are something you will need to continuously manage yourself.
A persistent, real-time state
The system continuously maps and syncs data from your different tools, so you're always looking at a consolidated, up-to-date picture of your fund's active state, not whatever was last updated manually.
Tools like OpenWiki, an open-source wiki-memory project by LangChain, show where this trend is heading for individuals. This tool connects to Gmail, Notion, Slack, and other platforms and keeps an up-to-date "wiki" made of markdown files so the agent you run can easily understand the state.
You can take inspiration from that if you want to build a system for a firm, but a personal wiki is just a starting point, although it remains a siloed solution.
Closed-loop
AI shouldn't just answer what you ask it. It should watch what matters and act on it. The gap between what a fund intends and what's actually happening is usually invisible until it's a problem. A closed-loop system closes that gap on its own: it monitors actuals against expectations, and when something drifts, it surfaces or handles it, without anyone having to ask.
That's the difference between using AI and running on it. Using AI means you open a tool and prompt it. Running on AI means the system is already watching, and it tells you before you would have thought to check.
To get there quickly, you can use any scheduling technique, combined with Claude Cowork or other Claw-type solutions, to watch the system and alert you to the things you care about.
Build on infrastructure, not around it
Becoming AI-native takes time, so start building the foundation now instead of waiting. The earlier you start, the faster your fund runs on it.
You're an investment firm, not a software company. Your edge is in finding, backing, and supporting elite founders, not in maintaining custom agent pipelines. To win in the AI-native era, build on top of a battle-tested infrastructure layer so your team can focus on what they do best: investing.