Stewart Butterfield on Product Craft, Utility Curves, and Why We Don’t Sell Saddles
Stewart Butterfield, co-founder of Flickr and Slack, in conversation with Lenny Rachitsky. A wide-ranging product masterclass covering utility curves, the comprehension-versus-friction distinction, the hyper-realistic work-like activities problem, Parkinson’s Law at organisational scale, when to pivot, and the origins of the ‘We Don’t Sell Saddles Here’ memo.
Key ideas
- Utility curves shape product strategy. Every product’s value follows an S-curve: a long flat period where most people extract little from it, a steep curve where a core segment gets enormous value, then another flat section. Understanding where your users sit on that curve — and who is on the steep part — determines everything about where to invest in the product.
- The challenge is comprehension, not friction. Removing friction is the right goal when users have high intent and know precisely what they want (Taylor Swift tickets). When users arrive with low intent and incomplete understanding, the challenge is not friction — it is comprehension. Building the right understanding of what the product is and what to do next matters far more than shaving clicks. Most product design is in this comprehension space.
- Hyper-realistic work-like activities (HRWLA). As organisations grow, the ratio of known-valuable work to total headcount inverts. People fill the gap with activities that are superficially identical to real work — deck reviews, pre-briefing meetings, alignment conversations — but create no output. The leader’s job is ensuring sufficient supply of known-valuable work to do, which is harder than it sounds.
- We Don’t Sell Saddles Here. Slack’s pre-launch internal memo, written when the company had eight people. Its argument: you are not responsible for building just the product but also, to a significant degree, for creating the market. Selling a saddle is not selling horseback riding. You must create the aspiration, not just the object.
- Pivoting requires cold rationality. Glitch (the game that became Slack) was shut down only after exhausting every non-ludicrous long-shot idea for making it commercially viable. The humiliation of admitting a pivot — to investors, employees, early users — causes founders to extend losing bets far longer than the expected-value calculation warrants. Distance from the emotional investment is required to make a rational call.
Comprehension vs. Friction
Stewart’s distinction is among the most useful reframes in the episode. Friction matters when intent is high and specific: if someone knows exactly what they want and wants it badly, obstacles genuinely block them. But most users arriving at a new product are just above the threshold of intent, with near-zero specificity. They do not know what the product is, what they are supposed to do, or what happens if they do it. No amount of friction reduction helps them.
The design question for most products is therefore: can someone look at this screen, understand what the product does, and understand what to do next? The failure mode is making users feel stupid — the cognitive cost of a decision they cannot understand is real and leaves a lasting emotional association with the product.
The UI implication: reducing the number of clicks is almost always the wrong goal. Reducing the amount of thinking required is the right one. A product where every step is trivially obvious but takes eight taps is better than one where two taps require navigating an incomprehensible options menu.
The Do Not Disturb Rollout
The DND feature implementation is Stewart’s example of investment paying off through careful rollout. Every Slack administrator was told weeks ahead of time. A system of layered defaults was designed: organisation-level default, admin override, user override, admin-can-re-override. The default was set because without it, most users would never turn on the feature at all — and the feature’s value depends on mass adoption.
The same logic applies to the @-channel mention redesign. When the product team brought back a pre-populated @mention in thread reply boxes (on the basis of a statistically marginal increase in thread length), Stewart rejected it. The analysis required feature flags, instrumentation, A/B test runs, dashboard creation, meeting scheduling, and rescheduled meetings — all to measure whether an average thread became 2.17 messages instead of 2.14. The decision was guaranteed to be a loser on expected value regardless of the statistical outcome.
Hyper-Realistic Work-Like Activities and Parkinson’s Law
Parkinson’s Law: work expands to fill the time available. Its deeper organisational form: the number of administrators grows independently of the work that requires administration. Stewart’s Slack example — 27 product managers, each wanting to hire a junior — illustrates the mechanism. The impulse is not malice but rational self-interest: headcount correlates with seniority, pay, and authority.
HRWLA (his term) is the output: people performing activities that look exactly like work — presentations in conference rooms, collaborative deck reviews, pre-briefing calls — but produce nothing. The paradox is that these activities are performed by smart, motivated people who genuinely believe they are working. The solution is not chiding; it is creating clarity about what constitutes known-valuable work and explicitly declining to fund activities outside that set.