1. Half-Life was designed in meetings
At one point in 1997, Valve had nearly completed Half-Life. The monsters were cool, and the technology was groundbreaking, but the game wasn't fun.
Ken Birdwell’s account of what happened next (from 1999!) is an excellent, occasionally uncomfortable read. He describes how a small group assembled the promising bits into one playable level. Everyone played this level and then immediately understood what needed to be done.
Then they had a lot of meetings: six hours a day of them, four days a week, for five months. In each meeting room were engineers, a writer, an animator, a level designer: the people responsible for making the actual thing, designing it together.
Longtime readers will know that I have often advocated for the ritual clearing of calendars, but it's worth remembering that in a lot of cases, the meetings are the work and the meetings can indeed be good. This story shows exactly what that looks like: make decisions about game, watch people play game, resist urge to explain anything to playtesters. Basically do demos and eliminate all distance between you and the product. (Imagine applying that rule to a strategy presentation.)
There's a bit in here about how someone might contribute nothing for two or three meetings, then become the main contributor for the rest of the week; a pretty good argument against measuring the usefulness of a person one meeting at a time.
2. You could add balconies
These three apartment buildings in Bordeaux contain 530 homes. Lacaton & Vassal, working with Frédéric Druot and Christophe Hutin, transformed them by adding winter gardens and balconies. They kept the principal structure, stairs and floors. The savings helped pay for more space, light and possibilities for the people living there. Balconies like this are also better for the climate emergency we're in today for the rest of time.
We're broadly taught that transformation should prefer reduction to addition, but this project shows us that important stuff is sometimes missing.
A team might have good relationships and useful expertise but spend half its week waiting for another department. Giving it access to a missing capability could change its working life considerably. You might not need to dissolve it, retitle it, or make everybody apply for their jobs. Let's see the balcony proposal before the demolition proposal.
3. Some of those saved hours never existed
METR has a note on AI productivity that distinguishes doing your old work faster from doing a different mix of work now that certain things are cheap. Imagine you generate an elaborate interactive dashboard in an afternoon. Building it manually would have taken three weeks. Have you saved three weeks?
Well. Previously, you might have just made a spreadsheet.
The dashboard could be much better! Perhaps people can finally understand something they couldn't see before! But the three weeks were never in the budget, and the value of the improvement needs its own accounting. METR calls the extreme version of this a "Cadillac Task": work undertaken because AI has made it cheap, with a much bigger apparent speedup than gain in value.
Which connects over to recent thoughts about project intake. Once we can make almost anything, prioritization has to accommodate a new category: things that are worth doing, but would never have been worth doing at their old price. I expect to enjoy quite a few of these, and we should still be careful adding up the hours.
4. The spreadsheet person gets a bigger workshop
Ink & Switch's essay on malleable software starts with an ordinary freedom: people rearrange their kitchens and workshops to suit themselves. Software rarely offers the same latitude.
Back to Ink & Switch: their ambition is for people to adapt tools where they use them, with a gradual path from using something to changing it. They also take communal creation seriously, which is nice.
Think about the person at work who knows Excel. Someone shows up with a problem, they make a thing, colleagues try it, and it improves. Over time, the spreadsheet contains a detailed understanding of how that particular group operates, including the exceptions nobody remembered to mention in the requirements. Give that person more powerful tools and the person who understands the work can improve the equipment.
That person will also, someday, go on vacation, and others will need to understand what they built. I’d love to see companies develop this as a supported practice, with time and help attached. Suddenly everyone has to become good (or teach their agents to become good) at documentation...
5. The restaurants with no menus
The New York Public Library has a wonderful collection of old menus. Much of it exists because Miss Frank E. Buttolph spent years asking restaurants for them. She wrote letters, placed advertisements, and pursued a collection that grew to roughly 25,000 menus during her involvement. The library's account of her is worth reading too; apparently she sometimes sent submissions back for failing to meet her standards.
Cartoonist Ben Katchor's research into dairy restaurants offers another angle into the collection, pointing out that many contemporary working-class Jewish restaurants were missing. Why? Some didn’t have printed menus. The waiter recited the specials, and the regulars knew what they wanted. You can have thousands of beautifully preserved records and still miss a whole way of doing things because it didn't produce the kind of object you collect. Perhaps there's a "the map is not the territory" lesson in there for those of us building world models.
Anyway, enjoy. The typography alone is worth it!
The best things you’ve missed since the last Five Things: Eskom made room for maintenance and got dramatically more reliable. UPS reorganized around the physical network. Meta tried shrinking conventional teams into 3–5-person AI-native pods. Joshua Gans has a neat model for how AI can redesign a job without automating any of it. Unilever’s end-to-end divisions turned out to be companies in waiting. And Target cut capacity from the center and bought more of it at the edge (some of the numbers moved in the right direction).
