Three Weeks at KeyReply: From Firefighting to Parallel Thinking
Notes from three weeks as an AI PM intern at a startup — leadership churn, hopping across products, 1:1s from CEO to interns, and reflections on parallel work in the AI era.
It’s been three weeks since I started my internship at KeyReply, and honestly, I haven’t found a clear direction yet. Within these three short weeks, the Chief Scientist who hired me and the head of the product team both left, one after the other. And I’ve ended up something like a firefighter — going wherever I’m needed. This rhythm of work is very different from how I worked in school: one demands that you plan and push many tasks forward in parallel, while the other has you stay focused on a single subject for weeks on end.
Is this kind of work inevitable at a startup? I don’t think it’s entirely so. Most interns who arrived a few months before me have stayed within the single domain of AI QA testing, with little lateral expansion. For me, by contrast, I’ve already touched nearly every product the company has — flagship products, research products, prototypes, and even concept-stage products that had just finished initial client conversations. Looking back, perhaps it’s the timing of my arrival that gave me this turbulent rookie phase.
Honestly, an experience like this is both a kind of mental strain — because it stops you from focusing fully on the lifecycle of any single product — and a good window for quickly building a sense of the whole company, and for training your ability to work in parallel. How to grab the key points in the middle of chaos, and how to keep a steady mindset against a messy calendar — that has been the first lesson I learned after joining.
In these three weeks, I also had 1:1 chats with about five colleagues, ranging from the CEO down to fellow interns. From peers, I drew on their hard-earned experience and reflections. From my direct manager, I gave concrete shape to my previously fuzzy understanding of the product role, and through our conversations gradually clarified an SOP for how the work should run. From the C-suite, I learned how to analyze and solve problems from the company’s or the client’s point of view, and how to fully leverage existing resources to maximize impact.
I think, in the age of AI, handling multiple work threads at the same time will gradually become the norm. If Claude Code or OpenClaw can act as the central hub that orchestrates different sub-agents, then in work, the human is that central hub. From this, I believe that in the current era, the ability to quickly recognize a task, break it down, and push multiple sub-tasks forward in parallel isn’t just a question for AI itself — it’s a question for the people who use AI.