Stewart Butterfield

The pattern they actually represent

Butterfield shipped the internal tool the dead project had already built. He did it twice, which is what makes it a pattern rather than a rescue.

Flickr came out of Ludicorp, the company he co-founded in 2002 to make a browser game called Game Neverending. The game was shelved. Flickr launched on 10 February 2004, assembled from tooling originally created for that game, and Yahoo bought Ludicorp about thirteen months later.

The year after he left Yahoo, he ran it again. Tiny Speck was founded in 2009 with three other Flickr veterans and raised more than $15m to build Glitch — a browser-based multiplayer game about crafting and gathering, built deliberately without combat. Glitch launched publicly in September 2011 and went back into beta that November. About forty people worked on it, spread across North America. To run themselves, they had built their own messaging system on top of IRC, with file upload, searchable archives and automated notifications. It had replaced their email. Nobody was paying for it. Nobody had been told to use it.

On 15 November 2012 he closed Glitch in public. The same announcement said the team had developed messaging technology with applications outside the gaming world, and that a smaller core team would work on new products. The ending and the next commitment were the same sentence. The servers went dark on 9 December, and a prototype of the surviving thing was running about a month later.

So the move is not a change of direction. The vague version of this story is the useless one. The move is an inventory. A failing project is a factory that has been producing two things: the product, which does not work, and everything that had to be built so the product could be attempted at all.

The blockage it speaks to

Read this one if the numbers already answered your question and you are still funding it.

Notice what this reader is not: confused. The evidence arrived a while ago. They are stuck because stopping converts a long stretch of their working life into a write-off, and every week they continue makes it bigger. It compounds rather than resolving.

Sunk cost is not really an inability to stop. It is an accounting error about where the value sits. The stuck founder is running a ledger with one line on it — the product — so closing the product writes the entire spend to zero, and nobody volunteers for that. The instinct underneath is sound: real work was done, and it did produce something.

Butterfield did not answer that with resolve. He changed the ledger. Years of building a game had produced assets that were not the game, and shipping them recovered the spend through a different object. The years were not wasted. They had been capitalised in the wrong place.

That is also why he is the right study here, and a poor one for a founder frozen between carrying on and stopping. He was not frozen; he decided quickly, twice. What the record hands you is narrower than courage: the asset was the tooling, and it already had daily users who were the wrong customers.

Three moves you can steal

Inventory what the dead project already built that works. Not a reflection — a list, in one sitting. Every internal tool, script, dataset, integration and working relationship it produced. Two columns per row: who used this without being told to, and what breaks for them if it disappears on Friday. It is an evidence task, not a courage task, and you can finish a list on a bad day.

Make the ending and the next commitment a single announcement. Do not send the ending on its own. In the same message, to the same people, name the surviving asset and who works on it next. What holds a founder inside a dead project is rarely the project. It is the gap on the far side of stopping, in front of everyone who watched. Naming the successor closes that gap before it opens.

Force one friendly team to switch fully, and turn the old thing off. Take the strongest row from your inventory and get one team outside your company using it as their only option for two weeks. Tiny Speck did this to themselves first. A trial with a fallback measures nothing, because the fallback absorbs every failure.

Where the pattern breaks

The by-product move is told as a decision. It also ran on a balance sheet. Tiny Speck had raised more than $15m for Glitch, and when the game closed Butterfield offered the remaining capital back to his investors; they declined and funded the next attempt instead. The four who then built the messaging tool were Flickr veterans who had sold a company together.

So the precondition is not insight. It is a funded, intact, senior team that survives the shutdown. If your dead project leaves you with neither, the honest version is smaller: ship the surviving asset as an artefact, not as a second company.

What to do this week

Run the inventory in one sitting. Then take the top row to one person outside your company and describe it to them. Then ask them to describe it to a second person while you listen.

Butterfield has said that if you cannot explain what you are doing to someone such that they can go and reliably explain it to a third person, it is probably hopeless. The test costs one coffee.

Listen for what survives the second-hand version. If it comes back unrecognisable, you do not have a product yet — you have a preference, and you found that out this week instead of one quarter from now. If it comes back intact, you are looking at the row that recovers the spend.