By the end of December 1973, the three men aboard Skylab could not keep up with their schedule. Gerald Carr, Edward Gibson and William Pogue were rookies, the third and last crew of America’s first space station, on a stay stretched from 56 days to 84. Planners had skipped the usual easy first weeks and scheduled them as if they had already adjusted. Their requests to lighten the load, NASA’s history office says, went unheeded for several weeks.
Near the end of December 30, Mission Control sent up a five-page summary of the crew’s work. Then, as the station passed over three tracking stations in a row, the crew talked with the capsule communicator, Richard Truly, for ten minutes at each one. Thirty minutes of radio, against weeks of friction over the schedule.
A 1980 Harvard Business School case taught a different version of that week: the crew had turned off the radio and refused to talk to Mission Control, and the strike story is still retold. NASA’s transcripts don’t support it.
What they record is a conversation. Carr later called it the first sensitivity session in space, because both sides aired their grievances.
Rules for a day in orbit
The recommendations from those three passes included a protected day off every ten days, fewer tasks around sleep, time between jobs to clean up and set up, and exercise in one block.
“I think it did a world of good,” Carr told Truly. By NASA’s account, there were no more incidents over the timeline.
Nothing on that list settles who was right. Each item is a rule about how the crew’s day gets divided, and until then the day had been divided on the ground. Seen through Tuckman’s model, that is the question worth asking when a new team fights.
When a new team starts fighting, ask whether anyone agreed how the work gets split.
Four rhymes from a pile of papers
The model itself came out of the Navy. Bruce Tuckman’s first job was at the Naval Medical Research Institute, studying small groups “as the US Navy prepared for a future of small crew vessels and stations.” His boss, Irwin Altman, handed him 50 articles on how groups develop and asked whether he could make anything of them.
Tuckman found four stages: testing each other out, then conflict, then cohesion, then people settling into working roles. He named them forming, storming, norming, performing. In 1977, with Mary Ann Jensen, he added adjourning, “a fifth stage for which a perfect rhyme could not be found.”
Read through the model, Skylab’s bad weeks look like storming. The model can’t tell you the crew had to pass through them. What it can do is point you at a question: a plan written on the ground had met the work in orbit, and for weeks the crew’s requests to change it went nowhere. New roommates are polite for a month; then someone leaves a pan in the sink, and the house finally gets its rules.
Thinner than the rhyme
By Tuckman’s own account, many of his 50 articles were “psychoanalytic studies of therapy or T-groups,” the training groups of strangers that the NTL Institute pioneered in the late 1940s. NTL’s history says efforts to run them inside companies mostly failed, because strangers lack “the multitude of ties, authority, and competitive relationships inherent in organizational life.” By 1977, only one of 22 newer studies had set out to test the model directly. Tuckman’s own guess about its fame: “quotability may be the key to success.”
Then the researcher Connie Gersick observed every meeting of eight project teams. She found no universal series of stages. Some teams began in harmony, others in storms. Each settled on an approach in its first meeting and kept it for half its time. Then every team changed course, “precisely halfway between its first meeting and its official deadline,” dropping old habits and making dramatic progress.
NASA puts Skylab’s radio conference at roughly the midpoint of its 84-day mission. One flight proves nothing about midpoints. Still, a curious place for the argument to land.
Two people, one launch page
Say you have four people and eight weeks to launch a product page. By week three, your designer and your analyst argue in every meeting over who has the final say on the copy. The easy reading is that you hired badly, so you settle it yourself.
Tuckman’s model suggests asking something first: was it ever agreed who decides the copy? If not, you could let them finish and write down what they agree: who decides, and by when. Gersick adds a date. Book a meeting for the end of week four to look at the whole approach again, and invite the manager who asked for the page. A team that has already had its argument may have more to bring to it.
The crew that forgets
With an AI agent, only one step of this carries over: stating the working rules plainly and saving them where the next session will find them.
Anthropic’s documentation for its coding agent says: “Each Claude Code session begins with a fresh context window.” What carries over is what’s written down, such as a CLAUDE.md file. The guide suggests adding to CLAUDE.md when “Claude makes the same mistake a second time.”
An agent making the same mistake twice is not a team working out its relationships. Nobody is testing anybody, and no trust is being built or lost. The earlier correction may simply not be in front of it.
Even then, the docs say Claude treats those files “as context, not enforced configuration.” For a rule that must never be broken, they point to a hook, a check that runs whatever the agent decides.
Skylab’s new rules took three radio passes to agree. An agent’s rules, once saved, will be in front of the next session. Whether it follows them is still yours to check.