Arista Networks

Arista had a lot of smart people in the first few years who laid a ton of "just right" groundwork. Many, but not all, of the early architecture decisions have stood the test of time.

The old-timers are also quite sharp.

Arista's onboarding is excellent and a great way for inexperienced people to quickly gain experience with a large, complex codebase.

The biggest con is that very little of what made Arista great five years ago is still true today.

Arista has consistently been following the next big pot of money placed before them by their biggest customers every quarter for years. In the past, this meant there were lots of exciting projects to take part in. In recent times, these sorts of projects have become rarer, and the technical debt left behind from previous mad rushes to make deadlines means the codebase is hard to work with.

There is limited appetite for paying down that technical debt. As a consequence, good projects are few and far between, with most projects being time-pressured death marches fighting against the code.

The tooling and infrastructure has a solid, old core, but most newer tooling built on top is either resume-driven development or half-finished implementations.

Projects complete more out of inertia from the pre-IPO employees who grew up with the codebase, with support from new grad energy. Management input has essentially no bearing on the success or failure of a project. Project selection is poor, and though it used to be true that you could switch between teams and projects with ease, that hasn't been the case for years.

There is no real avenue for advancement or professional growth beyond an intermediate level; the team you start in is likely to be the team you quit from.

All the senior managers are long-time Arista employees, and nearly all of them came up from the development ranks. Several joined Arista straight out of school and have no experience elsewhere. In most cases, managers are picked based primarily on tenure within the area, with leadership ability a distant concern.

As a rule, all the senior managers used to be decent coders but are now more or less incompetent at their jobs of management.

Direction from middle management is more or less absent. Innovation is supposed to be grassroots-driven, but there is no energy left after the constant fighting with tools, release bug triage, customer support, and, often a single real day a week, working on your assigned projects.

Consequently, what innovation occurs is small in scale and mostly related to skunkworks tooling improvements and refactoring. Product innovation primarily comes from customer requests (solve this problem, implement this standard, fulfill this RFE without implementing some old, massive standard suite), ASIC vendor products (vendor X releases a chip intended for a switch fitting in niche Y), and acquisitions.

Replace all directors but Simon. Staff 20% of all developers to refactoring projects to dig out of the technical debt hole. Make it a priority of every manager to have a plan for the professional growth of their reports.