Bee Partners has organized its investment thesis around vectors since roughly 2016, and around three specific vectors since roughly 2022: Human-Machine Interaction (HMI), Machine-to-Machine Learning (M2ML), and Biological Machines (BioM). The framework was useful. It gave us a shared vocabulary, structured how we explained the firm to LPs, and created a taxonomy for the deals we were seeing.
It also stopped fitting.
In this post, we retire the framework. Not out of embarrassment, but because frameworks serve until they don’t, and this one had run its course. A second post, coming soon, will lay out what replaced it: four futures we believe in.
How the vectors served us
When we introduced the three-vector framework, it did something important: it forced discipline. Every deal we considered had to connect to a legible thesis arc. Could we articulate why this company sat at the intersection of machines and meaningful human leverage? If not, we passed or sharpened the argument until we could.
The vectors also gave LPs a handle. Rather than saying “we back technical founders in AI-adjacent spaces” (which tells you almost nothing), we could point to specific domains and explain the underlying logic for each. HMI captured the shift in how humans and machines interacted at the interface layer. M2ML captured the compounding of intelligence across machine networks. BioM captured the convergence of biological systems with computational design.
For three years, those three arcs covered most of what we were seeing. Then they didn’t.
Where the framework strained
At our annual meeting last October, we were still presenting all three vectors. Kira framed BioM as “bio-adjacent” from the stage: a careful, honest description of where Bee’s actual bets had landed. The portfolio companies in that cluster were closer to lab automation, AI-enabled tooling for biological research, and process optimization than to biological machines in the original sense. We had become more interested in the enabling layer than in the biological systems themselves.
We left the AGM with the sense that the framework had stopped fitting. Not for BioM alone, but across the board.
The honest tell was the agentic workflow deals. A wave of compelling pre-seed companies was building agentic infrastructure, enablement tooling, and orchestration layers for AI systems. Most were obviously fundable on our terms. But fitting them into the three vectors required mental gymnastics: they weren’t pure HMI plays. They only sat in M2ML if you squinted at the data network effects. The categories were right-shaped for 2022; they were increasingly wrong-shaped for 2024.
When a rubric requires that much interpretation to apply, the rubric is the problem.
The precedent for evolving
This is not the first time Bee has reshuffled its thesis labels.
Our 2021 Vector-Driven Analysis whitepaper enumerated a wider taxonomy: Market Networks, Applied Data Vector, M2M Edge Intelligence, HCI (Human-Computer Interaction, the precursor to HMI), Synthetic Biology, and an industry-vector cluster. By roughly 2022, those had consolidated to three: HMI, M2ML, BioM. The broader categories had compressed as the field clarified.
We wrote this directly in that whitepaper:
“And while Bee Partners has yet to abandon a Vector in its entirety, it is certainly reasonable to think that a core thesis or technology we are now investing in may not be the same in three years’ time. A fundamental tenet of Vector-driven analysis must be its ability to evolve. Otherwise, it becomes the antithesis of what is required to effectively evaluate innovation at the technology frontier.”
And:
“By nature, Vectors are always accelerating towards an unknown horizon. If a Vector does evolve, it is often because that horizon ultimately looks different than originally surmised, supported by new or tangential technologies.”
We wrote those sentences in 2021. We meant them. Holding onto the three vectors past their useful life would have been the actual betrayal of the framework’s founding logic.
Late January 2025: the effective deprecation
By the end of January 2025, the three-vector framing had been replaced in our LP-facing materials. The narrative organized itself around something different: four futures we believe in, named for specific bets about how the world of AI-enabled systems would unfold. The vectors didn’t appear.
That was the effective deprecation. Not a formal announcement. Not a team meeting with a whiteboard. The new narrative existed; the old one didn’t appear in the new materials. That’s usually how thesis evolution works at the frontier: the new framing earns the space it occupies, and the old framing quietly becomes past tense.
We sat with that shift through the first half of 2025. Q1 gave us time to stress-test it: did the four futures hold under pressure from new deals, from portfolio dynamics, from conversations with co-investors? They did. We stopped reaching for the old vocabulary.
What didn’t change
The companies remain. HMI, M2ML, and BioM weren’t wrong theses; they were the right theses for a specific period of machine intelligence development. The portfolio companies that landed under those labels still represent exactly what Bee backs: technical founders at the liminal moment, building at the frontier.
On the individual portfolio company pages from earlier funds, you’ll still see vector labels in the company metadata. Treat those as records of how we framed each bet at the time of investment, not as active claims about the firm’s organizing thesis today. The framework retires; the per-bet history stays where it is.
The underlying conviction hasn’t moved. Machines have won; the question has always been which founders will build the machines that matter. That logic predates the three vectors. It will outlast whatever framework we use to describe it.
What changed is the rubric: how we organize and explain the bet to ourselves and to the outside world. Rubrics are instruments. A good rubric creates clarity and forces discipline. When it starts creating confusion and requiring workarounds, you replace it.
Where this goes next
The four futures we believe in get their own treatment in a separate post. It will explain each future directly, with portfolio signals and the underlying thesis for each.
This post is narrower: the three vectors served Bee Partners well, and the AI Supercycle revealed that the category lines had shifted in ways the original framework couldn’t accommodate. We said four years ago that vectors must be capable of evolving. This is what that looks like.
The door on the three-vector framework is closed. What comes next is on our What We Believe page and in the pillar essay to follow.