Skip to content

Conway’s Law

Melvin Conway is a compiler developer and systems designer, who is well known for the eponymous Conway’s Law. Various phrasings exist of that, and one popular is

Organizations which design systems are constrained to produce designs which are copies of the communication structures of these organizations.

The original paper and an introductory paragraph can be found on his website. It’s worth reading, because there are more useful insights to be found in the original writeup.

So what does this even mean? Can you give examples from your current or previous work environments?

What Conway says is that communication across personal and organisational boundaries is not free. He phrases that in the context of design processes by a large committee, which divides itself into subgroups and then delegates. This process already constrains the choices the whole group and the subgroups can make, which in turn will shape the outcome of the resulting design.

Thus the life cycle of a system design effort proceeds through the following general stages.

  1. Drawing of boundaries according to the ground rules.

  2. Choice of a preliminary system concept.

  3. Organization of the design activity and delegation of tasks according to that concept.

  4. Coordination among delegated tasks.

  5. Consolidation of subdesigns into a single design.

The subdivision and delegation phase creates boundaries of communication between groups. The Coordination phase among delegated tasks and the Consolidation phase of the subdesigns costs additional communication overhead from the various people and groups involved. Avoiding this resistance and energy expenditure is shaping the structure of the resulting solution to the point where that structure will mirror the organisational structure.

There are several corollary observations in his paper:

Conway makes interesting observations regarding overpopulation of design groups and how this has consequences for the design.

He points to Parkinson’s Law and states »As long as the manager’s prestige and power are tied to the size of his budget, he will be motivated to expand his organization.« This leads to large pools of people.

These people need to be used, or as Conway writes »A manager knows that he will be vulnerable to the charge of mismanagement if he misses his schedule without having applied all his resources.«

But large pools of people need to be subdivided into smaller groups to be able to work usefully, and that creates administrative boundaries which will shape the outcome of the design. »This knowledge creates a strong pressure on the initial designer who might prefer to wrestle with the design rather than fragment it by delegation, but he is made to feel that the cost of risk is too high to take the chance. Therefore, he is forced to delegate in order to bring more resources to bear.« in the words of Conway.

Flat hierarchies are important. In Conways world view of 1968, he correctly states that »the number of possible communication paths in an organization is approximately half the square of the number of people in the organization. Even in a moderately small organization it becomes necessary to restrict communication in order that people can get some “work” done« Modern corporate communication structures get away with more direct, less hierarchical communication structures than we had fifty years ago, and that is indeed a great source of progress, because it make it cheaper to cut through administrative boundaries and lowers cross-division communication cost.

Conway’s Law also works backwards for operations organisations: In order to operate a system successfully, you must create an organisational structure that matches the design of the system that you are trying to operate.

You can observe Conway’s law in motion around you every day:

  • A developer finds a bug in a program that is supplied by an external vendor. Will the developer have the energy to put their own progress on hold, create a bug report, drive the bug and repair discussions in the remote organisation and then integrate the fix?That is not usually what happens. The developer will instead integrate a workaround into their code and may or may not take it upon themselves to report the bug upstream.
  • A company has a security organisation that is a separate division from the development organisation. The rollout of a new corporate security policy and tooling makes it impossible to install unmanaged plugins in the managed main browser. Some of these plugins are necessary or helpful for the development organisation to do their job. Will the development organisation individually or collectively engage in a discussion with the security organisation in order to be exempted from the misguided security attempt?That is not usually what happens. Instead somebody will point out that workarounds exist (Run an unmanaged browser, run a browser unmanaged remotely on a box not controlled by the security organisation and mirror the screen onto the developers desktop, or others). There is a strong incentive to not point out these workarounds to the security organisation, because shutting down these workarounds is a reasonable risk to assume and the development organisation will protect their ability to be able to do their job.
  • A security bug exists in a piece of software from an outside vendor. The outside vendor does not accept any bug reports, or does not honor them, if they are reported by random people that are not support customers of their organisation. Will the finder of the bug take it upon itself to convince the vendor to accept the bug report even in the absence of a support contract?That is not usually what happens. Instead the security bug might be ignored or will be sold for money to organisations that care about knowing about security bugs that nobody else has knowledge of, or it will be anonymously publicised in order to shame the vendor into fixing it. The third way is actually the best outcome here.
  • A developer needs access to the interface of the service another part of the company provides. While the web interface of the service is accessible, the underlying database is not directly accessible. Access is being denied in as many words: “Access denied.” instead of pointing out how to actually apply for accesa and which permissions are needed. Will the developer take it upon itself to find ways to properly access the data in question.In the observed example that is not what happened. Instead a screen scraping, HTML parsing solution has been implemented.

What can you do?

Organisations that put up barriers, in the name of departmental structure or security, need to understand that the majority of access attempts are genuine and potentially legal. An efficient and agile organisation will make itself accessible and will provide clear instructions how to communicate with each failed or denied communication attempt.


  • Never print “Access denied”, but explain how to get access.
  • Never turn away bug reports, but make an effort to distinguish between bug reports and support requests.
  • Never establish security controls without running them in logging instead of enforcement mode at first. Use that log data. Differentiate groups of users and understand their legit needs.
  • Establish a preference for trailing controls instead of chokepoints. That is, log and gobble, parse and investigate the log whenever possible instead of preventing access whenever you can get away with that.
    • Do this especially around the development part of your organisation.If the security branch of an organisation annoys the development part of an organisation they just created a formidable in-house adversary that will try to subvert their efforts through literal or creative interpretation, unexpected workarounds or atypical user behaviour.

What else can you do?

Break down organisational barriers where they turn out to become development obstacles. That is for example, if the security organisation and the development organisation are not aligned enough for the purpose of the company, try to mix and integrate the two. DevOps and later SecOps are attempts of actually doing this.

Many corporate shuffles and re-organisations are actually attempts of corporate structures to align themselves better and bring down communication costs. Understanding the reshuffle and trying to match that up with the real or perceived problem that this reorganisation is trying to solve can be very instructional: in judging the management and communications skills of the management involved, and in recognising the perceived or actual problems the organisation has.

Always be aware than when you change an organisation, the software system they created before is a mirror of their old organisational structure and the old structure of the software system that organisation created will now work against the new organisational structure.

Allow for time and effort to mitigate that. Expect delays.

Published inComputer ScienceErklärbär

One Comment

  1. rajo

    Thanks for sharing! Given the current organizational restructuring I’m part of, it will be interesting to watch if and how these effects can be observed (or hopefully: prevented).

Leave a Reply

Your email address will not be published. Required fields are marked *