Reflections on my work trial
Earlier this year in January, I left my job to explore new horizons. With the rapid rise of LLMs, I realized that both our workflows and the products we build are fundamentally changing. Wanting to step back and create tools I wished already existed, I began building several side projects using models like Claude and Codex.
Along the way, I observed that while frontier models excel at solving problems, they can easily stray off course and invent their own context without the right guidance. Conversely, providing too much context to the AI agents can backfire by poisoning prompts with irrelevant information. While trying to solve this for myself, I discovered that SageOx was tackling very similar challenges around team context. I reached out to connect, and one conversation led to another. Soon after, two weeks ago, I entered the interview process for an engineering position at SageOx, my first "interview" in nearly four years.
The experience was far from typical. Rather than following the usual sequence of a recruiter chat, a hiring manager screening, and a series of system design and coding panels, I sent an email outlining my background and why I'd be a strong fit. After reviewing my profile and side projects, the team invited me straight to a work trial. This unconventional format completely sets SageOx apart from traditional interviewing, which I've never particularly enjoyed.
The process
I almost hesitate to call it a "process." For me, that lack of process was a huge draw. It thrives on pure agency and independence, though I can see how it might feel daunting to others. There's no playbook, no rigid machine. I spent 4 days embedded with the team as a regular member, with full access to all systems and zero gatekeeping. It was just diving into real work and shipping something meaningful, which gave me the most authentic view possible of what it's actually like to work here.
The most impressive aspect is getting to know the team far better than any standard one-hour interview could ever allow. It creates a mutual win-win for both the candidate and the company. It easily surpasses a typical take-home assignment where you tackle problems in isolation, which isn't representative of true team collaboration.
Overall, I had a fantastic experience working alongside everyone on the SageOx team, building something both deeply engaging and valuable to the product.
The project
SageOx is reimagining human-agent collaboration from first principles, exploring how the rise of LLMs and agentic workflows will transform team dynamics.
Right as I started my work trial, Block launched Buzz, a novel tool for human-agent collaboration. The team was thrilled about its potential, and so was I. I had already set up Buzz for my personal agents before even starting this, so I dove in to explore it further.
Not surprisingly, the team at SageOx uses SageOx - the product - all day long to dramatically improve their own efficiency. SageOx, the product, supercharges teams and their AI agents by keeping everyone aligned on ongoing tasks, standup updates, and key decisions. It delivers the right context to both humans and agents exactly when needed.
SageOx and Buzz seemed like the right primitive to explore how they can work together. So for my work trial project, we focused on integrating SageOx's team knowledge into Buzz agents, giving them direct access to full organizational context, across the team and beyond work that happens strictly in Buzz channels. Within a few days, I quickly ramped up and deployed a hosted agent with integrated SageOx context. This allowed anyone on the team to ask natural questions in Buzz like, "Hey, what's everyone working on?" and see a summary of the activities happening within Buzz, across coding sessions in multiple harnesses, MCP tools, and human conversations in meetings.
Below is what the team was working on and the agent automatically used ox CLI to detect it.
Reflections
One particularly compelling observation from my experience was how effectively SageOx tools accelerated my onboarding compared to any previous process. By checking the SageOx Dashboard, I could easily see what teammates were focused on and stay up to date on relevant discussions.
The murmur feature stood out as a true game changer, automatically capturing coding context from teammates and surfacing it whenever relevant.
Working alongside the SageOx team was an absolute blast, which ultimately made my decision to join them full-time an easy one. As team collaboration continues to evolve, especially in an era where each person orchestrates multiple AI agents, the complexity of collaboration will naturally scale. I firmly believe SageOx is addressing these exact challenges head-on to help humans and agents collaborate far more efficiently.
PS, We're hiring :-)
