05. Practice

The mythical man-month, and what its author later revised

He withdrew the advice to build a pilot and discard it, in print, twenty years later, and stood by the rest.

For a separate operational view of time, ownership and team activity, see this Monitask guide.

mythical-man-month.src
royce1970.pdf fig. 2 — "this concept is risky"
# the diagram everyone copied

1975The claim the book is known for

Adding people to a late project makes it later. The reasoning is in two parts and both are usually dropped when the line is quoted.

New arrivals must be trained, by the people who are already behind, which subtracts from the work immediately and adds to it only later. And the paths of communication between people grow roughly as the square of their number, so each addition costs every existing member a little more of their attention.

The condition attached to it

The claim is about tasks that cannot be cleanly partitioned. Where work divides into genuinely independent pieces with narrow interfaces, adding people works, and the book says so.

Quoting the line as though it forbade ever adding anybody is the commonest misuse, and it is used to justify refusing help on work that would divide perfectly well.

What the number actually is

For a group of three there are three connections. For eight there are twenty-eight. For fifteen, a hundred and five.

Nobody maintains all of them, which is the point: beyond a small size the group stops being one group, and either it is deliberately divided with defined interfaces or it divides itself informally and the interfaces are undocumented.

The proposal nobody remembers

The book's constructive suggestion was a small team organised around one person doing the primary work, with the others in defined support roles, on the model of an operating theatre.

It was not widely adopted, partly because it is hierarchical in a way the profession has become less comfortable with, and partly because it requires the central person to be reliably excellent. Its underlying observation, that conceptual coherence comes from few minds rather than many, survives in every argument about architectural ownership.

1995The retraction

The anniversary edition contains a chapter in which the author revisits the book and states plainly that one of its recommendations was wrong.

The advice to build a pilot version and expect to discard it assumed that the second version would be built to a specification settled by the first. He concluded that incremental development, growing a working system by successive refinement, is the better model, and that the earlier advice had been superseded.

Why the retraction matters more than the law

It is one of very few cases in this field where a widely cited author publicly revised a widely cited recommendation, in print, twenty years later, with reasons.

The recommendation he withdrew is nevertheless still quoted, sometimes in the same paragraph as the law he stood by, which illustrates the citation behaviour this document's part on practice keeps returning to: what travels is the memorable sentence, not the book.

What he did not retract

The central claim, the argument about conceptual integrity, and the essay on the absence of a single large remedy were all restated. So the book is not discredited by the retraction; it is a case of an author distinguishing between the parts of his own work that held and the part that did not.

Which is a considerably better model of how to treat one's own published views than the alternative, and is rarer.

The unit in the title

The phrase objects to a unit of accounting. Estimating work in person-months treats people and months as interchangeable, which is true only where the work divides without communication.

That the unit remains in universal use, including by people who quote the book, is a small illustration of how much easier it is to adopt a slogan than to change a spreadsheet column.

What the square actually predicts

Not that large teams cannot work. That the cost of holding one conversation across all of them grows faster than the team, so beyond a small size the structure must change: subdivision, defined interfaces, written decisions rather than remembered ones.

Every organisation discovers this and most discover it late, at the point where the informal arrangement has already been overwhelmed. The prediction is fifty years old and the discovery is made fresh, which is the pattern the part on repeated numbers closes on.

three people, three pathseight people, twenty-eightWhat grows when a team grows
FigureThree people with three connections and eight with twenty-eight, which is what actually grows when a team grows.

Why it is quoted so much more than it is read

The title contains its own summary, the central claim fits in one sentence, and the sentence is useful in an argument somebody is already having.

That combination is what makes an idea portable, and this document has now described the same mechanism for a diagram, a ratio and a phrase. The next part is entirely about what happens to a number when it acquires that property.

present dayHow the law is misused in practice

Two directions. It is used by teams to refuse help they need, on work that would divide. And it is used by managers to justify understaffing indefinitely, since if adding people cannot help then the current number is by definition sufficient.

The book supports neither. Its actual advice is to staff realistically at the start, because the expensive moment is the addition made late under pressure, not the addition itself.

What we cannot verify

The book and its later edition are published and the retraction can be read directly. The underlying experience was one organisation's, building one operating system, in the nineteen-sixties, which is a single case however influential. No study we would rely on has measured the effect of adding people to late projects in general, because the experiment cannot be run.

In short

  1. Training subtracts immediately and adds later; communication paths grow as the square.
  2. The claim is about work that cannot be partitioned, and the book says so.
  3. Three people have three connections; fifteen have a hundred and five.
  4. The constructive proposal was a small team around one principal, and it was not adopted.
  5. The author retracted the advice to build a pilot and discard it, in print, twenty years later.
  6. He stood by the rest, which is a better model than either defending or disowning everything.

also in Practice

Next.

further context

For a primary or institutional reference, see the GNU licence reference.

Every claim here carries the source it came from.

The source and its year sit beside the sentence they support. A secondary account is marked as one, and where the record is unclear the entry says so rather than choosing the better story.