I recently joined a panel at DLA Piper's Private Credit Academy alongside Eric Chang (Claira), Daniel Liechtenstein (Hypercore), and Sai Uppuluri (Houlihan Lokey) to talk about how AI is transforming private credit. The conversation was supposed to be about technology, but it ended up being mostly about people.

Bottoms-up adoption wins

There's a popular impulse to form an "AI committee" - a top-down body that evaluates tools and sets strategy. I'm pretty anti-AI committee. Find your analysts. Give them unfettered access to tools as fast as possible. They'll tell you which ones are working really quickly. The signal comes from the bottom, not the boardroom.

Run your proof of concept right

For products where people generally understand the workflow being automated, run the POC in parallel with the current work stream. Use a skinny set of data but complete the workflow end to end. Yes, people are doing double the work for a short period - that's how you actually see the value.

For newer categories of tools - the ones defining something you couldn't do before - let the vendor define what the POC looks like. I've seen funds hand an AI underwriting tool a stack of materials and say "produce your own IC memo." That's not a fair test.

Your junior team is already being transformed

The ramp into junior roles is changing dramatically, and every ninety days as the models update. The role is becoming less about number crunching and more about judgment. Start thinking of yourself as a manager of agents - a single person managing a swarm of agents doing tasks simultaneously.

About a year ago, my engineering team told me they don't write code anymore. My first reaction was probably the same as yours. But what they're actually doing is reviewing code. Our newest engineers talk to the agent, sit back, and then review every single line. They're learning - just in a completely different way than pair programming.

If you're an analyst at a private credit fund right now, you probably won't be spreading financials manually in a year or two. But you should still understand the fundamentals. The skill to develop is judgment - and doing it sooner than your predecessors had to.

Build vs. buy: the car engine analogy

Think of AI as the engine of a car. The heuristic: will this product fundamentally get better over time as models improve, or will it become more redundant? Every Anthropic / Open AI marketing release kills a set of companies. If the tool you're evaluating is on that kill list, you should probably be worried. But Salesforce, a core system of record, isn't going anywhere. When Anthropic rolls out a new model, it probably makes Salesforce's agent better, not obsolete. Is the product the car, or is it just a thin wrapper around the engine?

What comes next

AI is going to let you ask questions that no one's been able to ask before - because the data was locked in email threads, SharePoint folders, and models that walked out the door when someone left the firm. The people who will thrive aren't the ones who can crunch numbers the fastest. They're the ones who know which questions to ask.

Watch the full video here: https://vimeo.com/1174174978/6c66a44c94?fl=pl&fe=cm