Boiled Tofu

Some people just don't like tofu. That's real, and I'm not going to try to talk anyone out of it.

But most of the people I've heard say it have eaten tofu exactly one way: a pale cube, boiled, sitting in a puddle of its own water. That isn't tofu. That's the worst available thing to do to tofu. And on the strength of that one plate, a whole category got filed under "not for me," permanently.

Coffee gets it too. Burnt gas-station sludge at 6am, case closed. Jazz gets it. Poetry gets it worst of all, from everyone who last encountered it being made to strip-mine a poem for symbolism in a high school classroom and never went back.

One encounter ends up speaking for an entire category, and it usually wasn't a fair representative. From the inside that feels like a preference. It's closer to a very small sample that somebody else picked for you.

But it's not always that simple. Sometimes there's no better version to go find — the thing is neutral, and the setting is the only thing that makes it good or bad.

Sometimes the thing is fine and the room is wrong

Reach for the thing almost everyone is sure is simply bad: a wildfire tearing through a forest. Catastrophe, plainly — evacuated towns, lost homes, half the West under smoke every summer. But a lot of what makes those fires so destructive is that we spent a century putting every one of them out. Decades of aggressive suppression let dead wood and undergrowth stack up — in some places to five times the historic level — so when something finally catches, it has far more to burn.^[1] The remedy the Forest Service and Indigenous fire practitioners keep arriving at is more fire, not less: deliberate, controlled burns that clear the fuel before it can feed a monster. A Stanford study published in Science found that a prescribed burn drops the chance of an extreme wildfire in that area by roughly 90 percent, with the benefit lasting more than a decade.^[2] The element we treat as the enemy is, applied on purpose, most of what stands between a forest and catastrophe.

Water runs the same way. Glen Canyon Dam gave forty million people dependable water and power, and in doing so cut off the seasonal floods that once hauled sediment down the Colorado and rebuilt the Grand Canyon's beaches and sandbars.^[3] So the government now floods the canyon deliberately — multi-day "high-flow experiments" that open the dam's bypass tubes to imitate the floods it erased, moving hundreds of thousands of tons of sand back onto the banks.^[4] Nobody welcomes a flood, but removing them turned out to be the actual damage.

Two different mistakes, one lazy conclusion

"I tried it and I hated it" collapses two errors that have nothing to do with each other:

People treat both as evidence about the thing itself. Neither one is.

The MVP is where we do this to ourselves

On Peter Yang's Behind the Craft, Sanchan Saxena — who ran product at Airbnb — walks through a small thought experiment about how Brian Chesky thinks. Say you want to build Airbnb Lounges: somewhere guests can sit for a few hours between a 7am arrival and a 3pm check-in. How do you test it? The trained PM instinct is the scrappy version. Small space, cheap coffee, free wifi, something you can scale later. Chesky's instinct, per Sanchan, runs the other way: pick one location and make it the best lounge in the neighborhood. Real barista. Good chairs. Air conditioning that works. Then figure out what scales.

Sit with what would've happened on the first path. You open the room with the cheap coffee, people are lukewarm, and you write in the postmortem that guests don't want a lounge. Idea dead. Except you never tested the idea. You tested a thin version of it and then held the concept responsible. A mediocre MVP sells your product's potential short.^[5]

Worth being clear that this was a hypothetical he used to make a point, not a real lounge Airbnb built and quietly buried. Which is arguably worse. It means the failure can happen entirely in your head, in a planning meeting, before anything gets built at all. You picture the cheap version, feel underwhelmed, and kill it.

Sometimes the bad version actually ships

The Apple Newton. Pen-based handheld computing was a sound idea — it turned into the defining product category of our lifetime. Apple shipped it in 1993 with handwriting recognition that mostly produced nonsense, reliably enough that The Simpsons got a joke out of it: "Beat up Martin" comes out as "Eat up Martha."^[6] It sold roughly 50,000 units in its first four months^[7] against internal hopes of a million in the first year.^[8] Nobody was rejecting handheld computers. They were rejecting a broken one and assuming that settled it. Three years later the PalmPilot did the same idea properly and the line from there runs straight to the phone in your pocket.

Apple Maps in 2012. Owning your own mapping stack instead of letting Google decide which features you get was strategically obvious. The launch warped landscapes, invented roads, and at one point routed drivers across a live taxiway at an airport in Alaska. Tim Cook published an apology.^[9] The executive running it was gone within weeks. The product is genuinely good now and I still know people who won't open it because of one bad afternoon fourteen years ago.

New Coke, for the inverse. Coca-Cola ran roughly 190,000 blind taste tests and the sweeter formula won.^[10] But a sip favors sweetness in a way a whole can doesn't, the test stripped out everything people actually feel about Coke, and nobody thought to ask how they'd take it if the new formula replaced the old one. A confident answer, properly collected, completely wrong. Same disease as the tofu cube, pointed at your own research.

Which brings us back to the backlog

Once you're listening for it, this is most of what gets said in a product org.

Users hate this feature. Do they, or did they meet the buggy half-shipped version, form an opinion, and never come back? When someone tells me "we tried that and users hated it," the only useful reply is: which version, and under what conditions? Shipping something rough once and watching it flop isn't a finding.

Users hate change. They don't. They hate change dropped on them with no warning, no path back, and no explanation. The same feature that gets torched in a surprise rollout gets adopted fine when it shows up with some context around it. Nothing about the feature improved in between.

That research method doesn't work. Someone watched one usability session run badly — leading questions, wrong participants, a facilitator steering toward the answer they wanted — and wrote off the method. Sushi, judged from the refrigerated tray at the grocery store.

And the one that ought to sting, because it's the same logic aimed back at us: dark patterns are persuasive design in the wrong room. A nudge that helps someone finish what they came to do is good work. Urgency, defaults, friction — the identical mechanisms — pointed at somebody trying to cancel a subscription is something else entirely. I've made this argument before about retention: make cancellation painful and the quarterly number looks great while the relationship rots. The mechanism isn't the problem.

What to actually do about it

Not "give everything a second chance." That's exhausting and often unwarranted. Boiled tofu really is inedible and some ideas really are bad.

The narrower discipline: before an experience becomes a verdict, work out which of three things you're actually holding.

Most of the strong opinions in the room are the second or third wearing the costume of the first. Somebody had one bad cup of coffee, the team absorbed it as fact, and now there's a roadmap decision resting on a boiled tofu cube.

So when you hear "we tried it and it didn't work" — including from yourself — don't take the verdict. Ask which version, in what context, compared to what. Usually it turns out nobody tested the thing at all.