The loop around the model.
A good model is only the beginning. Funds get useful AI by defining a bounded workflow, its trusted context and the checks that return judgement to the team.
Hi folks,
A good model is only the beginning. For a fund, the useful work is in the loop around it: clear ownership, trusted context and checks that return judgement to the team.
The one thing
For a fund, the AI work is designing the loop
- I wrote about engineering loops this week: give an agent a clear goal, the right context, a way to check its work and a stopping point for human judgement.
- A deal-screening loop might check a new teaser against the mandate, flag what is missing and return a first view. An analyst still decides whether it deserves time.
- The loop improves as the team learns. You do the work manually, see where it breaks, then make the checks and context more useful.
What this means for a fund:
- Start with one workflow, one owner and one definition of done.
- Define the checks: which sources count, when it must escalate and how an exception is recorded.
- Ask: what repeatable work would we trust an agent to do, and what proof would we need?
In the mix
- OpenAI is turning agent deployment into a product (OpenAI Presence)
- Presence starts with a specific job, controlled access and explicit policies. Teams test the agent against edge cases before it reaches users.
- Why it matters: the useful agent is not a chatbot with broad access. It is a maintained operating loop around a valuable workflow.
- Open models are becoming a real option, not an automatic cheap option (Associated Press)
- Moonshot's Kimi K3 is a serious open-weight model competing with leading US systems.
- Why it matters: model choice is now a risk and operating decision. “Open” does not remove the need for hosting, security, evaluation or a fallback plan.
Readings
If you want to double click into things this week I’d recommend:
- The future is millions of specialised models: A thoughtful 20VC conversation with Fireworks AI founder and CEO Lin Qiao on:
- Custom models,
- Falling token costs,
- Why the answer may be a portfolio of models rather than one winner.
Fireworks AI just announced a $1.5 billion raise, betting on custom models as the future.
Levercon builds the AI operating system for credit funds: we connect your data, deploy agents across origination and monitoring, and run the repetitive work end to end.
Book a strategy call.
See where AI fits in your fund.
Walk through Levercon's custom AI operating system to see how it can benefit your fund.