Small Teams, Different Minds, and AI
I am interested in teams that can notice a problem, test an idea, and learn from the result quickly. Small teams can make that easier. Their size alone does not guarantee it.
I live with ADHD. In my own work, I notice details and patterns quickly and often keep several technical threads active at once. Sensory overload is also part of everyday life. Both belong in the picture.
That is my experience, not a description of everyone with ADHD or a claim that neurodivergent people work in the same way. I do not want a diagnosis to become another template people are expected to fit. What interests me is how a team makes room for different ways of thinking while remaining accountable for its work.
Make room for the person and a record for the team
I like following a problem across boundaries. In the inference project, that meant moving between GPU kernels, model state, benchmark design, and a Rust serving layer. That range fits how I approach engineering. It also makes a written record important: another person needs to understand which explanation survived testing and which one was abandoned.
The kernel investigation records hypotheses that failed as well as changes that helped. That is a concrete practice I value. A team should be able to revisit the reasoning without depending on whoever happens to remember the experiment.
AI assistance belongs inside that process. It can help produce candidate implementations and explanations, but the team still needs to check them. Generating more possibilities does not remove the need to decide which question matters.
Small teams have useful conditions, and real limits
A small team can keep context close to the people doing the work. Fewer handoffs can mean a shorter path from an observation to an experiment. That advantage is strongest when the scope is bounded, the team has the necessary expertise, and someone clearly owns the decision.
It weakens when one person becomes responsible for everything. Production operations, specialist review, support, and sustained maintenance require capacity as well as initiative. A small team can accumulate technical debt and make poor decisions just as a large one can.
Large organizations can give a team access to equipment, expertise, and independent checks it could not provide alone. The useful question is whether the team can use those resources and make decisions within clear boundaries. A small, empowered group inside a large organization may have both advantages.
Keep the controls that improve the work
I want less unnecessary waiting. I also want security and compliance requirements understood early enough to influence the design. An independent review that finds an unsafe tool permission is useful feedback. An unclear approval chain that repeats the same discussion without a decision is a process problem.
For me, a good working environment combines ownership, written decisions, room to question assumptions, and access to people who can challenge the result. It should accommodate individual needs without assuming that every person with the same diagnosis needs the same arrangement.
My bet is on teams that learn quickly and can show their work. AI can support that ambition. Turning it into reliable engineering still requires attention, expertise, and care for the people doing it.