Q1 2025 was a hard quarter for private markets. Tariffs sparked broad risk-off behavior across institutional capital, and exit timelines extended further than anyone wanted. The macro friction was real and we didn’t ignore it.
Inside Bee, the energy ran in a different direction. From the Q1 LP letter, sent this morning:
“Inside Bee, we’re seeing a surge of energy and creativity, perhaps sparked by ‘vibe coding’ (building without a blueprint, experimenting without permission). It’s taken hold among our team as well: everyone’s working on something.”
What the letter couldn’t fully convey is what that energy produced.
The first LP letter we built (partially) with AI
The Q1 2025 letter is the first Bee Partners LP letter assembled predominantly with AI assistance. We’ve been trying, but our previous efforts didn’t come close. The portfolio company updates section was drafted by AI, then reviewed, edited, and refined by us. The first pass, the structure, the synthesis of what each company accomplished in the quarter, came from the machine.
That is a real threshold; not a gesture toward AI-native operations, but a shift in how the firm runs.
We didn’t announce it in the letter. We just did it. The letter went out this morning, and it reads like a Bee letter because it is.
What we built in Q1
The portfolio-updates experiment didn’t happen in isolation. It was the output of a quarter spent going deep on AI-native workflows across every part of the firm.
The internal work covered both tooling and process. We built a LangChain-and-RAG chatbot trained on Bee’s own materials: investment memos, thesis documents, founder correspondence. The intent was a shared intelligence layer, useful to the team for deal prep and useful to LPs as a way to access firm context on demand. We built Python scripts for data transformation and standardization across the portfolio. We shipped a React-based self-serve intake form for founders, replacing a workflow that required manual handling on our side.
Individual workflows evolved in parallel. Multi-project Claude and ChatGPT configurations became the standard environment for deal prep, pass memos, and thesis logic. We produced a 15-minute AI-generated audio episode about the newest portfolio using NotebookLM, which told us something real about how much structured knowledge we’d accumulated and how accessible it had become. Alejandra returned from NVIDIA’s GTC with sharper perspective on what the compute and capital stack for the next decade actually looks like. David Pitman, our Venture Partner focused on AI infrastructure, published a widely circulated piece on scaling AI systems that stretched our own internal thinking.
We also cut tools. [redacted] was trialed for diligence and not retained. A genuinely AI-native firm tests, evaluates, and drops tools as fast as it adopts them. The stack shifts every quarter. What matters is the judgment about which tools earn their place in the workflow.
Cursor became the prototyping environment. Granola handled async recording and note capture. Lovable and Gumloop pulled founder sentiment from Reddit and Discord. Tools that would have required significant engineering time to configure two years ago now get stood up in an afternoon.
The result: more speed, more structure, more time allocated to the work that requires judgment rather than execution.
This is what “AI-first” means in practice
We’re starting to describe Bee as an AI-first firm. That phrase appears in what we believe. It deserves to mean something specific.
Every firm has adopted AI tools by now. What’s different at Bee is that the operational infrastructure of the firm, from LP communications to deal analysis to internal knowledge management, runs on AI-native workflows as a baseline rather than as an experiment.
The Q1 LP letter is the clearest proof point we have. Assembling a quarterly letter that covers each company across our active portfolio, synthesizing each update into a coherent picture of what the quarter meant: that task used to take weeks. The AI-assisted version took a fraction of the time and produced something we were comfortable signing.
We are still the decision-makers. The judgment calls on what to include, what to sharpen, what to cut: those are ours. But the first pass belonged to the machine, and the machine did well.
…and, we’re off!