Cobra Effect · People and organisations
Conway’s law
Systems end up shaped like the teams that build them.
7 cards, read aloud in 2:37, with a test and sources.
A company has four teams. It asks them to build one program.
Melvin Conway, a programmer, noticed what came out. The program had four parts. Not because the problem had four parts. Because the company did.
Conway wrote it up in 1967. A business journal turned it down.
The paper was called How Do Committees Invent? The Harvard Business Review said he had not proved his thesis. A computer magazine printed it the next year. Any organisation that designs a system, he wrote, will produce a design that copies the organisation’s communication structure.
Fred Brooks read it, and named it Conway’s law.
In 1975 Brooks put it in his book about why software projects run late. The organisation chart, he saw, is the first draft of the architecture. Whether anyone meant it to be or not.
Why it happens. To build two parts that fit, the two builders have to talk.
If they sit together, the joint is smooth. If they are in different buildings with different bosses, the joint is a contract, a form, a queue. The system has a seam wherever the people do. It cannot be otherwise. A design is a record of who talked to whom.
In 2008 a Harvard Business School study measured it.
Alan MacCormack and colleagues compared software built by loose open source groups with the same kinds of product built inside firms. The open source code was more modular. The corporate code was more tangled, like the corporations. The mirror was real.
Which gives you a lever most people never pull.
If the system copies the organisation, then design the organisation first. Want three clean parts? Make three teams that own them, and give the teams reasons to talk across the joins. Engineers call it the inverse Conway manoeuvre. Managers call it the organisation chart.
Next time a system seems oddly shaped, find the chart.
The duplicated feature is two teams that didn’t know about each other. The clumsy handoff is a boss who doesn’t talk to a boss. Ask who would have had to talk for this to be simple. Then ask why they didn’t.
Sources
- How Do Committees Invent?, Melvin Conway, 1968. The original paper, hosted on Conway’s own site, with his account of the rejection by the Harvard Business Review. Short and still sharp.
- Team Topologies, Matthew Skelton and Manuel Pais, 2019. The practical handbook for the inverse manoeuvre. How to arrange teams so that the system you want is the one that falls out.
- Conway’s law, Wikipedia. The statement, the variants, the Harvard Business School evidence, and the ways companies have tried to use it on purpose.
Nearby ideas
- Brooks’s law. Adding people to a late project makes it later.
- Dunbar’s number. Why groups stop working the old way once they pass about 150.
- Path dependence. Why early choices stay locked in long after better ones appear.
- The Hawthorne effect. Being watched changes what people do, and muddies what you measure.
- The bystander effect. The more people who see a problem, the less each one acts.
- The fundamental attribution error. We blame people’s character and overlook their situation.
- The Abilene paradox. How a group can agree to something that nobody wants.
- The Pygmalion effect. What we expect of people quietly shapes what they become.