Prerequisites

This guide assumes:

  • Familiarity with Conway’s Law. Read What Is Conway’s Law? first if “communication structure shapes system structure” sounds new.
  • You work on or with a platform team. This means a team that owns shared infrastructure consumed by multiple product teams.
  • Visibility into platform PR queues, product-team workflows, and meeting calendars. Without these signals, you cannot measure friction.
  • Influence over platform priorities, or a partner who has it. The fixes below require changing what the platform team does, not what product teams ask for.

Measure the friction budget

Platform teams own shared infrastructure that many product teams depend on. Heavy edit volume across multiple teams is normal. Slow shipping and burned-out engineers are not.

Measure the cost first. Count three things over the last two sprints:

  • Platform PRs whose review wait exceeded three days.
  • Times a product team shipped a workaround instead of using the platform.
  • Hours platform engineers spent in coordination meetings per week.

Those three numbers are the friction budget. Anything above one workaround per sprint, or more than six coordination hours per platform engineer per week, means friction is winning.

A “workaround” means a product team shipped functionality that duplicates, bypasses, or wraps the platform: a local cache where the platform exposes a query, a hand-rolled retry where the platform has a client library, a private copy of the platform’s schema. Count those, not minor config overrides.

Apply the three levers

If any number is over budget, apply the inverse Conway maneuver: make it cheaper for product teams to use the platform. Reduce the communication friction.

Three levers work:

  1. Self-service documentation. Write guides that let product teams use the platform without asking questions. Fewer questions mean fewer meetings, which means cheaper communication.
  2. Clear ownership and SLAs. Tell product teams who to ask, when to expect answers, and what the platform guarantees. A minimal SLA fits in one paragraph: “PRs from product teams: first review within 24 business hours. Outages: 15-minute response, 4-hour mitigation target. Roadmap requests: triaged weekly, answered within two weeks.” Publish this somewhere product teams can find it without asking.
  3. Reduce coupling. Find places where product teams must coordinate with the platform to ship. Remove those places. The simpler the interface, the cheaper crossing it becomes.

Pick one lever and commit. Running all three in parallel splits attention and produces no measurable change.

Verify

Re-measure the same three numbers after three months. The lever worked if:

  • Platform PRs over three-day wait dropped by at least half.
  • Workarounds per sprint dropped to zero.
  • Coordination hours per platform engineer dropped under four per week.

If only one number moved, the lever addressed one friction axis. Pick a different lever for the others. If none moved, the lever was the wrong choice; pick again from the remaining two.

Troubleshooting

The commit graph still shows congestion three months later

The lever you picked was the wrong one. Look at which of the three friction numbers moved and which did not. If PR wait time dropped but coordination hours stayed high, documentation cut question volume but did not reduce coupling. Pick a different lever and try again.

Product teams still escalate platform requests

Ownership and SLAs are unclear or unenforced. Publish them, then enforce them. If platform leadership cannot honor an SLA, the SLA is fiction and product teams will keep escalating through whoever they know.

Conway’s Law does not seem to apply

A few cases resist the law’s predictions:

  • Very small teams (fewer than ten engineers) where everyone already talks to everyone.
  • Strong shared tooling that flattens team boundaries (a monorepo with aggressive review automation and shared linting).
  • Regulated industries where audit boundaries force the architecture.

Confirm one of these applies before concluding the law is wrong. In most other cases, the law is operating; you have not measured the right thing.

Where to go next

References