AI adoption in an investment firm usually moves through three levels: individual prompting, personal automations, and firm-wide AI infrastructure. The distinction is not how advanced the model is. It is whether context, workflows, and accountability are shared by the firm or trapped with an individual.
This is a qualitative framework from recurring conversations with private-market investors, not a market-size survey. Its purpose is to help a team name its current operating model and decide what to improve next.
| Level | Where AI lives | What works | What breaks first |
|---|---|---|---|
| Basic user | Individual chat and research tools | Faster drafts and summaries | Context has to be recopied and work is not shared |
| Tech-savvy intermediate | Personal scripts, skills, and tool connections | Repetitive tasks become faster | Knowledge and maintenance stay with the builder |
| Power user | A custom firm-level stack | Shared workflows and richer context | Reliability, permissions, and upkeep become an internal product problem |
The three levels of VC AI adoption
The basic user
You use consumer-layer apps such as ChatGPT, Claude, or Gemini for summarization, basic analysis, and research. The tools are useful, but the operator has to copy context from emails, notes, and the CRM into each prompt. The result belongs to the person who asked, not to the firm.
This level is often the right place to start. The next step is not “use a more powerful model.” It is to choose one repeatable workflow where retrieving context, producing an output, and checking the result can be made easier for the whole team.
The tech-savvy intermediate
You have a deeper understanding of the AI landscape. You may be building agent skills, connecting tools through MCP, or writing scripts for a recurring task. This is where a team begins to see real operational leverage: a workflow can become faster, more consistent, and less manual.
The catch is ownership. When each person has their own context, prompts, and scripts, the firm has not yet created a shared capability. A useful test: if the person who built the workflow is unavailable for a week, can someone else understand, run, and improve it?
The power user
You custom-build an AI layer, sometimes with a dedicated technical owner. You want a proprietary company brain, a knowledge layer, and centralized workflows. At this level, the questions are no longer only about capability. They are about uptime, permissions, data handling, evaluation, and who owns the system as models and tools change.
That is not an argument against building. It is a reason to price the operating burden honestly. The NIST AI Risk Management Framework frames deployed AI as an ongoing governance, measurement, and management responsibility—not a one-time implementation.
The AI-adjacent trap
Each level can create value. The trap is treating a tool connection or a successful prototype as an operating system. A firm is still AI-adjacent when its workflows are hard to find, depend on one person, lack the context needed to act safely, or cannot be evaluated after they run.
Anthropic's guidance on effective agents makes a similar practical point: start with the simplest pattern that can succeed, then add complexity only where the task, tools, feedback loops, and human oversight call for it.
The ceiling of the advanced DIY fund
Building in-house comes with a ceiling when building software is not the fund's core business. Maintaining a reliable foundation while adapting to changing models and methods takes more than a successful demo. It requires clear owners, permissions that mirror the source systems, feedback on output quality, and a process for improving workflows without breaking them.
Basic apps are useful but isolated. Personal automations are useful but siloed. Custom stacks can be powerful but come with an operating burden. The goal is not to skip levels. It is to build shared, governed workflows deliberately.
It takes a shift in perspective: from using AI tools to running on AI-native architecture. In our next piece, we break down what that actually means for a fund's daily operations.