NetBox Labs Innovation & Technology Culture

What tools support your day-to-day work?

My day-to-day workflow starts with a team meeting in Zoom, recorded by Fathom, processed by Claude with meeting notes automatically added to Notion and tasks added to Linear. I proceed to develop with Claude Code in Cmux (though I still use VS Code to read and review code) and manage any asynchronous conversations through Slack. I keep a personal wiki, managed by Claude, with summaries of useful articles, progress on my work projects and other internal notes. We have also just launched an internal agents platform that includes an agent-native data warehouse of our internal business knowledge (Grid) together with a reasoning agent that we interact with in Slack (Flynn). Many of my workflows are now managed by Flynn and Grid gives me quick and searchable access to best practices other engineers are developing in their repositories as well as updates across the company.

How does your team experiment?

Being the AI team, we are constantly experimenting with how best to drive agentic development practices for our team and the company. Everyone is figuring this out and there are new ideas almost every day. We document our learnings and mistakes so they are shared across the company. Failures are first-class findings and we aren't afraid to share them. I run an augmented engineering guild which meets every week to share new ideas and try new tools as they come out. Team members are encouraged to explore and report back to the group. Recent findings range from directory-specific prompts performing ten times better than one monolithic config file to a runaway test-driven development session that burned $250 in tokens. Both are now part of our shared playbook. Our internal agents platform has been one of our biggest experiments and has taught us a lot about designing and operating one, which will lead to future product development work for our customers. On a more structured note, all of our AI products are shipped with an evaluation suite. This allows us to confidently experiment with different model types, prompts and tools to ensure our customers get the best experience.

How does your company adapt to change?

We have the benefit of being nimble and adapting to change quickly. Which is fortunate since the AI landscape shifts every few weeks. Change isn't an event, it's a steady state and we build our processes around it. There are two dimensions here that I face on a regular basis: the products we build and the tools that power them. On the tools side, I learned the 'bitter lesson' of agent engineering first-hand, over-engineering a harness around one model's limitations only to have to back that work out when a better model arrived. At NetBox Labs, we now expect models to improve and keep evaluations in place so we can assess new models and deploy them quickly when they are released. On the product side, we are constantly listening to customers and the community and move quickly to get new products out there into our customers' hands and then iterate on their feedback. We have a publicly defined product lifecycle that allows us to release and adjust as the market shifts.