I've been reflecting recently on Wave's growth spurt in 2019-21. Most teams grew 2-4x a year for multiple years, and culture and effectiveness stayed remarkably strong compared to what I'd have expected (or heard of elsewhere).

Some thoughts on what might have helped: 1. We held a super high bar for integrity + mission alignment. This was huge—IMO the root cause of most (hard to fix) dysfunction is people optimizing for themselves and not the team, so ~full mission alignment does more than anything to keep orgs from breaking. (See also QT:)

https://twitter.com/benskuhn/status/1629965839171731456

A lot of talk about managing focuses on "decisionmaking": how to run decision meetings, who gets to sign off on what, how they flow up + down the hierarchy...

But IMO, management isn't (mainly) about decisions; it's about understanding and tweaking a complex system (of people). Most individual A-vs-B(-vs-etc) decisions don't matter much in some sense. First because most decisions are reversible, so low-stakes. Second because making a decision is often straightforward: think about the pros and cons; think about what you care about; take your best shot.

One thing I love about both Wave (past job) and Anthropic (current) is how much everyone who works here cares about the mission. I think it's easy to underrate how much of a difference this makes if you haven't worked in that kind of environment. Some ways it's different... (Sidebar: It's hard to write about this without sounding naive, because every startup pays lip service to having a mission. But IMO most times it's bogus—the mission isn't that credible or compelling, or they don't prioritize alignment when hiring as highly as Wave/Anthropic do.)

A thing I often find myself suggesting to new managers is to "exert more backpressure."

Backpressure is a concept from fluid dynamics (and distributed systems) meaning the way in which a system resists overload—e.g. by slowing down, dropping requests, or completely failing. (See: tedinski.com/2019/03/05/bac…)

You generally want to build systems that correctly handle downstream backpressure (i.e. propagate it upstream), and exert upstream backpressure in ways that are easy for the upstream to handle.

There's a really interesting disconnect between how most people talk about "flat hierarchies" with few/no managers (a joyful world of freedom and harmony) vs how people who actually experienced them talk about them (a bunch of sad people flailing around chaotically) A flat hierarchy does have one upside, which is that it makes it much less bad if your boss is bad, since they're spread too thin to spend a lot of time making you in particular miserable. I wonder how many flat-management proponents have only experienced bad managers.

I've been overseeing a few cost-reduction projects at Wave recently, which is kind of fun since you have a much more objective feedback loop for whether your estimation / prioritization was right.

Here's some advice for this type of work I've found myself repeating: 1. Use numbers everywhere! Sounds basic, but I hear people talk about "not very much money," "a lot of users," etc. surprisingly frequently. Get in the habit of replacing every vague quantity word like that with a real number.

A decent number of people replied to this with variants of "effective altruism meetups." I love a lot of EAs (actually this might be my answer as well) but EA meetups work worse than I'd hope because of the classic meetup problem: lots of people with bad conversational habits.

https://twitter.com/benskuhn/status/1550093800957251584

At big EA meetups, I'll often end up in an interesting conversation with someone, only to have it disrupted partway through by someone joining in and asking uninteresting questions, raising silly objections, "well-actually"-ing, etc.

I generally didn't get much out of school, but today I'm thinking about the single class that probably made my uber fancy high school pay for itself by making me a massively better writer.

A big part of that was the textbook, which was awesome and extremely underrated IMO: It was _Style_ by Williams: amazon.com/dp/0226899152

What I really appreciated about it was that, unlike most writing advice, it gave lots of explicit, concrete and actionable mental models for what made particular phrasings clear vs. confusing or "choppy" or "turgid." E.g.:

When I wrote benkuhn.net/11, I got a lot of comments like "I've done >1k useless 1:1s, so there's no way 1:1s can be consistently great."

Hilarious self-own aside, I realized I didn't give a good playbook for *how* to do good 1:1s. Here's mine (for folks on both sides): (Credibility: I have also prob done >1k 1:1s, and unlike the commenter above, my most consistent piece of feedback in 360° performance reviews is that I give super helpful 1:1s.)

I don't claim this is the best way to do 1:1s, but it's some stuff that's worked well for me.

FUN FACT: I just did another round of "what's making my Zoom calls stutter every 60s" and this time the culprit was... APPLE #!*$ING MAPS. That's right, Macs now come *preloaded* with software to ruin wifi latency :(

You can fix by revoking Maps' location access in Preferences: Previously:

https://twitter.com/benskuhn/status/952245158078279680

This time, Maps was requesting a location scan every 60s, which triggered a wifi network scan, which, as previously discussed, tanks your latency for a few seconds.

Sorry to everyone I've had connection issues with for the past 2 months!

If you've ever modeled a real decision on the "optimal" policy for the secretary problem (reject 37% of candidates, then select next one who's better than any so far), PSA: that's only "optimal" for a dumb objective function that was chosen specifically to have a clean solution!! That policy leaves you with a 37% chance of selecting the applicant who happens to be last (= a random applicant). Intuitively, you might think that's pretty bad; can it really be optimal?

People often claim you should only build software if it's a core competency of your business, and buy the rest. Increasingly, I disagree: it seems possible to execute much better than leading vendors even in non-core areas.

For e.g., copying data between databases: Wave wants to sync data from instances of Postgres, likely the #2 most widely-used relational database, to Snowflake, likely the #2 most popular data warehouse. This is about as close as you can get to a 100% commodity problem that ~every midsize company has.

People often say that school is good for kids because it "socializes" them, i.e. teaches them how to get along with people.

This doesn't really make sense to me, because in practice kids seem to mostly copy the terrible ways that other (equally un-socialized) kids behave. What was great for socialization for me: growing up in a co-op, where I had to get along with a bunch of unrelated adults. I learned a lot about how to not accidentally offend or annoy people, have reasonable conversations, compromise about stuff, etc.

Hooray, @juliadwise is writing again! Enthusiastically recommended: juliawise.net

Julia is super thoughtful, a wonderful parent, and a role models of balancing effective altruism with having a meaningful life in other ways. And a great writer too!

Some favorites: Experiences raising children in shared housing: juliawise.net/experiences-ra…

(I was raised in shared housing and it was awesome!)

I used to love programming because it had such short feedback cycles. I still do, but now I also think that those short feedback cycles often trap smart people in local optima.

If you're a senior engineer trying to have a bigger impact, consider whether this is holding you back! Programming is a very friendly learning environment: you write some code, test it almost immediately, and (at well-run orgs) ship it very quickly. It takes a few days to see your work's value.

Unfortunately, most other work has no such near-instant satisfaction!

Google Docs shipped checklists and internal links while I wasn't looking!!

Since it's also the fastest document editor by far, this is making me strongly consider switching work docs to it... Back when we migrated from Hackpad (wow that brings back memories!) to Quip, we also evaluated Docs as a contender, but lack of checklists were the main dealbreaker because our task management system was ~entirely based on checklists in docs.

Thinking about the time @PeterSinger gave a talk at my high school, which was a pivotal moment in my life. I used to say he convinced me to be altruistic, but now I think something else was more important:

That was the first time someone told me *I could do things that matter.* By default, everything you do at that age is basically fake. Nominally you're studying stuff that will help you later in the real world, but there's no actual feedback loop with the real world and a lot of the material is really obviously not the most important stuff for that.

A genre that I wish existed and doesn't seem to: "tacit knowledge case studies"

An important part of acquiring tacit knowledge in e.g. management, software design, etc. seems to be getting lots of training data—trying out a lot of ideas, seeing which ones go well/poorly + why. So, you should be able to really speed up tacit knowledge acquisition if, instead of observing all that training data first-hand, you can instead learn about it from books/articles/whatever. This should allow you to observe the "important parts" of many more datapoints, faster.

I'd guess that if you want something like this to actually happen, you need to treat it like any other attempt to bootstrap a network effect:
- Expect it to be *very* slow
- Expect to put in tons of energy upfront w/ no payoff
- Start with 1-2 people who hate status quo most

https://twitter.com/nickcammarata/status/1463683133606301699

Thank you for coming to my TED talk about how moving to a village is the same as running a money transfer company

Trying to articulate a difficulty I have with the effective altruism community:

One thing that I, and I think many others, value about EA is that it's an antidote to using motivated reasoning to justify whatever you're currently doing for non-altruistic reasons as high-impact. For this reason, people care a lot about what activities "count" as effective.

Periodically, people try to convince EAs that their "pet cause" is effective. In order to justify excluding those people, EA wants very objective standards for what "counts."

I'm hiring for a ton of high-leverage eng leadership roles that will work closely with me!
- Director of Eng (x3)
- Eng Manager, Database
- Head of Technical Recruiting
If interested, or you know a great candidate, DM me—you'll have a big impact on Wave + the world!

Details: First, why work at Wave?
1. Banking the unbanked has a huge impact: if we succeed we'll likely help 10m+ people escape extreme poverty
2. We're growing quickly, so you'll learn and grow a ton
3. It's the most beloved product I know of:

https://twitter.com/benskuhn/status/1437898431977111556