Back to Insights
8 min read

Why Engineering Teams Lose Velocity After 50 People (And What Actually Fixes It)

The playbooks that got you to 50 engineers will break at 100. Here's what we've seen work—and what doesn't—across a dozen scaling teams.

E
Elena Vasquez
March 15, 2024

Every fast-growing engineering org hits the same wall. Around 50 engineers, things start to feel different. Decisions that used to take a day now take a week. Features that should be simple keep getting blocked by other teams. Engineers start asking "who owns this?" more than they ask "how do we build this?"

This isn't a mystery. It's a predictable consequence of how organizations scale. But the solutions most companies reach for—more process, more meetings, more documentation—usually make things worse.

The Real Problem: Coordination Costs

When you have 10 engineers, everyone knows what everyone else is working on. You can make decisions in Slack. Code reviews happen fast because there's always someone available who has context.

At 50 engineers, that breaks. Not because people got worse at their jobs, but because the number of potential interactions grows quadratically. With 10 people, there are 45 possible pairs of people who might need to coordinate. With 50 people, there are 1,225. With 100, there are 4,950.

Most companies respond by adding "lightweight" coordination mechanisms: standups, sprint planning, architecture review boards. Each one seems reasonable in isolation. But they compound. Before you know it, engineers are spending 30% of their time in meetings.

What Actually Works

The companies that maintain velocity do three things differently:

First, they invest heavily in internal platforms before they think they need them. The goal isn't to build sophisticated infrastructure—it's to reduce the number of decisions engineers have to make. If deploying a new service requires a 2-hour meeting to coordinate with DevOps, you have a problem. If it requires filling out a YAML file and pushing to main, you don't.

Second, they make ownership extremely clear and extremely local. This doesn't mean strict code ownership that blocks collaboration. It means every system has a clear owner who can make decisions without escalation. When ownership is ambiguous, everything slows down.

Third—and this is the controversial one—they resist the urge to "align" teams too tightly. Some duplication is fine. Some inconsistency is fine. The cost of perfect consistency is usually more coordination, and coordination is expensive.

The Mistake Everyone Makes

The biggest mistake we see: hiring senior managers to "fix" the coordination problem. This almost never works.

The issue is that coordination costs are structural, not managerial. You can't manage your way out of a system with 1,225 potential coordination points. You have to redesign the system to have fewer.

This usually means accepting some things that feel uncomfortable: teams that operate more independently (even if they make different choices), systems that have more redundancy (even if it seems wasteful), and standards that are looser (even if it offends your inner architect).

A Concrete Example

One of our clients—a Series B fintech—was struggling with exactly this. Deploy times had gone from 30 minutes to 4 hours because every deploy now required manual approval from three different teams.

Their first instinct was to hire a "release manager" to coordinate deploys. We talked them out of it. Instead, we helped them build automated deployment pipelines with feature flags, so teams could deploy independently without affecting each other.

Deploy times dropped to 20 minutes. The "release manager" hire became unnecessary. More importantly, teams stopped waiting on each other.

The Uncomfortable Truth

Scaling engineering teams is mostly about accepting tradeoffs. You're not going to maintain the same velocity at 100 engineers that you had at 20. The question is whether you lose 20% or 60%.

Companies that try to maintain perfect coordination lose the most. Companies that embrace some chaos—while investing in the infrastructure to make chaos manageable—do much better.

The goal isn't to eliminate coordination costs. It's to make sure they grow linearly, not quadratically, as you scale.

Found this useful?

We write about engineering, product, and how to actually ship software. If you want to talk about any of this, reach out.

Start a Conversation