In the five weeks since Headless GPX opened Juniper Square's data model to AI agents, GPs have started building real things against it: investor tearsheets, org charts, centralized reporting that pulls from Juniper Square and other systems at once.
What's stood out is that customers are spread across three stages of the same journey:
- Some are glad Juniper Square is building in an AI-native direction but want to be shown where the value is and how to get there.
- Some are already self-sufficient, using the MCP for easier, well-defined use cases.
- Others are building more ambitious agentic workflows.
We expect every customer to move through all three stages eventually. And the ones starting now will outpace their peers.
But even among the most advanced builders, a pattern holds: the workflows people hand to AI first are reporting, data aggregation, synthesis, and drafting. Tasks where a wrong answer is easy to catch. The workflows people hesitate to hand over are the ones where the agent has to act, not just assist. It's the same reason ChatGPT and Claude found instant, widespread adoption while autonomous browser and computer-use agents have moved more cautiously: people trust AI faster when they're still the one clicking "send."
Most private markets operations work, however, consists of more action-oriented workflows. Realizing the full value of an AI agent means trusting it to act, not just assist. That trust doesn't come from the model. It comes from everything built around it: the deterministic rules that make the same inputs resolve the same way every time, the audit trail that can reconstruct any decision long after it was made, and the controls that stop an agent before it acts on a bad call and roll the action back when it does.
Building AI systems well is a genuinely new skill, and almost no GP has it yet. Every firm will need that expertise, whether they build it in-house or partner with someone who already has it. It's the conversation we're having with customers every day.
Compliance is one example of a set of AI use cases where this conversation is tangible. Anthropic's release of a KYC agent this spring as part of their Claude for Financial Services release put a real, usable tool in front of every compliance officer evaluating AI for the first time. It's a capable piece of cognition: reading messy documents, reasoning over what they mean. But it is not, on its own, a compliance program. Like a car is more than an engine, the tools entering this space handle the easy part and leave the harder work for someone else to build: the accountability, the audit trail, the judgment on the highest-stakes calls.
The promise of compliance automation is exciting for both LPs and GPs, and means faster onboarding, less back-and-forth, and lower cost without giving up rigor. But it requires a whole system, not just an Anthropic tool. The firms getting this right are the ones automating the entire program, accountability and audit trail included, not just the parts AI makes easy. Building the car, not just the engine.
Juniper Square has run compliance programs for our customers for years, and our teams are already doing this work with agents in production. We're sharing what we've learned so every GP can make this transition.
Read An engine is not a car.