AI & ON-CHAIN RESEARCH
Researching how AI agents work together.
Founder of QuorumDesk, an independent research lab.
My background, and what keeps me working on AI and on-chain systems.
I’m Jacob Segade, an Australian-born AI and on-chain researcher with Spanish roots. I grew up mostly in Australia and now travel back and forth to Valencia to spend time with family.
What draws me to AI is the way a system’s behaviour changes as you connect it to other systems. I’m especially interested in agents working together, each with its own role and only part of the information needed to complete a task.
That brings me to the architecture around the model: how work is divided, what gets remembered, and how agents check each other’s reasoning. I want to understand how those choices shape the result, and where they create room for error.
I like following a question into an experiment and seeing where it leads. A result becomes much more interesting when I can trace how it happened, test whether it holds up, and work out what to change next.
From QUT and peaq to independent research and the people I’ve learned from.
My route into research has taken shape through university, a remote internship and a lot of independent learning. The conversations along the way have mattered as much as the formal steps.
I studied at Queensland University of Technology from 2021 to 2024, then deferred to complete a remote internship at peaq.
Since then, I’ve continued studying AI and on-chain systems independently. I like getting a concept into a form I can test, whether that means narrowing down an agent’s task, changing the information it receives or deciding how to evaluate its response.
peaq builds blockchain infrastructure for the machine economy, helping robots and connected machines participate in economic activity.
My interests gradually took me beyond that focus on machines and physical infrastructure. I wanted more room to explore AI agents, shared state and the ways activity across a network could be checked.
I went on to pursue those questions independently. They now run through the work I’m developing at QuorumDesk.
peaq on XThrough informal meetings and virtual tutoring, I got to know many people who work, or have worked, at OpenAI and Google, along with others at Anthropic. Some of those conversations became ongoing relationships.
A few employees at OpenAI and Anthropic have been especially important to QuorumDesk. They helped me work through the original idea, refine its direction and feel confident enough to start the lab. Having people take the idea seriously and spend time helping me think it through meant a great deal.
We still talk every day about the ideas I’m working through and where the research is heading. They’re enthusiastic about the concept, keep up with the lab’s development and continue to offer feedback. It’s become a steady part of how I think through the work.
They’ve asked to remain anonymous, which I respect. These are personal relationships; they don’t represent an official affiliation with or endorsement from OpenAI, Anthropic or Google.
What I’m exploring at QuorumDesk, and the questions I’m still working through.
I started QuorumDesk to give these experiments a place to develop. It’s an independent AI research lab focused on multi-agent systems: networks of agents working through a task together.
The hive idea is what keeps drawing me back. Each agent can have a different role, its own context and an incomplete picture of the task. I want to understand how those pieces come together, and what gets lost or distorted as information moves between them.
I’m looking at the choices that shape a network: how tasks are assigned, which information agents share, and what they carry forward in memory. I want to know when these choices help agents reason together and when they make it harder to spot a mistake.
Evaluation is a large part of that question. If several agents agree, I still want to know how they reached the conclusion. The useful evidence is in whether they checked each other’s reasoning, brought in new information or simply passed the same error around.
I’m happy to run an experiment at almost any scale, but I keep coming back to the idea of a small town. It’s how I picture a network with enough agents for specialised roles and lasting relationships to emerge, while still being able to follow what’s happening between them.
As that network grows, I want to see what happens to its memory and coordination. More agents means more interactions to account for. I’m curious about the point where those interactions start getting in the way, and how a failure in one part of the network reaches another.
My interest in on-chain systems runs alongside the AI work. I want to explore whether a shared record can make it easier to follow what agents did, how decisions were reached and where each contribution came from.
The harder question is what that record tells us about the quality of the work. I’m also curious about how on-chain incentives might change the decisions agents make along the way.
For people asking similar questions, or coming at them from a different angle.
If you’re working on agent systems, evaluation or on-chain infrastructure, I’d enjoy hearing how you approach it. A question you haven’t resolved yet or an experiment that changed your thinking would be a good place to start.
You can follow the work through QuorumDesk’s website and X account as the lab develops.