Forming, storming, norming and performing
Bruce Tuckman’s 1965 model of group development: teams don’t start effective — they become effective by passing through named stages, each with a different failure mode. The sequence: forming (polite, tentative, goals unclear, everyone on best behavior) → storming (friction surfaces: roles, standards, who decides; conflicts that were suppressed in forming erupt) → norming (resolutions stick, working norms and trust form, “we” language appears) → performing (the team executes without negotiating itself each time; energy goes to the work, not the working). Tuckman added adjourning/mourning in 1977 for team endings.
The value isn’t the letters — it’s two claims underneath:
- Friction is a stage, not a bug. Storming feels like the team failing; the model reframes it as the team processing. Leaders who suppress conflict to keep things comfortable freeze the team in forming — permanent politeness, never norms.
- The stages cost time, every time. A team that has passed through them is an asset that a same-sized group of strangers is not. That’s the whole economic point of keeping teams together.
The vault’s 2026 uses
Two independent captures apply the model, from opposite directions:
- Hiring molecules, not atoms (ghuntley × Josh Wulf, captured): Stripe’s playbook — “three people who’ve already been through forming, storming, norming and performing together.” The acqui-hire economy’s logic: in a market where individuals are increasingly leveraged but coordination is the scarce input, a pre-formed team is the asset worth buying whole. Kin: What matters in AI world (the two-slice/pirate-architect pairs are molecules of two), AI native teams (teams survive as the unit of work — Anthropic: “two humans + agents is NOT a team”).
- Model-switching resets the ladder (same capture, ghuntley): “switching models is like firing a developer and hiring a new one: you go back through forming, storming, norming, and performing with your silicon partner every time. I stick with a model for a long time and re-evaluate my priors now and then.” The hidden switching cost of model churn — you don’t just re-learn the model’s quirks, you re-negotiate the collaboration norms. Counterpoint in the vault: the AI gateways role-alias pattern (public name = role, model swappable) exists precisely to make the client side stable across switches; ghuntley’s rule argues the operator side still pays the stage tax.
Honest limits
- The model is descriptive poetry from 1965, not a law — later research (e.g. teams-as-oscillating-systems) shows groups loop and regress, especially under membership change; stages are revisited, not graduation ceremonies.
- It says nothing about how fast a team moves through storming — which is the variable that actually matters, and what good facilitation buys.
- The stage tax is the durable part: whatever the shape, coordination norms are built, not bought, and every reset (new member, new model, reorg) rebuilds them.
Kin
- AI native teams — the org-design lane: what team identity means when members run agents
- What matters in AI world — molecules-of-2 as the scarce pairing; charisma (Titus Winters) as the norming accelerant
- Career development · Senior mindset — storming-handling as a senior skill