When I started at Amazon in 1998, there were 60 people in tech. Not 60 SDEs. 60 people total, including DBAs, SAs, TPMs, managers, etc. We were divided into two teams of roughly 30 people. Each team met every other week and all 60 met together in the alternate weeks.
-
-
Obviously, you do really need to have discussions about strategy, or software design, or whatever. Gathering different perspectives is important. Letting many people have a voice is important. You just need to be conscious of the cost and pay it when it's worth it.
Prikaži ovu nit -
It's so easy in these discussions to become attached to being "right," and to continue to argue your position even if the other position is just different, or if it's unknowable which one will prove to be right. This is beta - remember that it has N² impact on productivity.
Prikaži ovu nit -
Sometimes it is important to continue to argue. We make decisions every day that have lasting ramifications. But often it's really hard to know how the decisions will play out. What you do know is that time spent coming to consensus is time nobody is producing software.
Prikaži ovu nit -
For more-senior people, especially, it can be hard to let things go when you think the path being proposed isn't the best. But try to ask yourself whether paying the cost of consensus delivers more than letting the less-senior person try it their way. You might be surprised!
Prikaži ovu nit -
Amazon has grown its tech community by an average of over 30% per year for over 20 years, so I've had a lot of opportunity to observe this process. And my observation is that it's hard for everybody to adjust to growth, regardless of their position on the team.
Prikaži ovu nit -
Consider the least-senior member of the original personalization and community team. He was involved in all the strategy discussions for the combined team, got to weigh in on the software for both pieces - generally knew everything going on.
Prikaži ovu nit -
After the split, he is on one of the two teams, and is only participating in strategy discussions for that team. He probably isn't participating in the design discussions for the shared software. It feels like his world has shrunk.
Prikaži ovu nit -
In reality, the total complexity of the business goals and software being developed by the post-split team of six is commensurate with the original team. It's deeper, but it's not as broad or diverse, so it feels smaller.
Prikaži ovu nit -
On the other hand, consider the most-senior member of the original team. She maybe doesn't end up on either team, but rather carries responsibility for both. Her challenge is that she can no longer keep up with all the details on both teams to the same level.
Prikaži ovu nit -
She has to let go of decisions that she would have weighed in on if there were only six people total. Now that there are twelve on two teams, she has to figure out what she needs to pay attention to, and what should be completely delegated to people on the individual teams.
Prikaži ovu nit -
Complicating this all is that it's not really a discrete process where you were on a team of six and wake up the next day to two teams of twelve. It's a continuous process where teams grow one person at a time and then semi-split-but-not-really-until-it-it's-way-too-many-people.
Prikaži ovu nit -
I thought about this a lot when we were growing the QLDB team over the past year. I talked to the team about how they might find themselves feeling left out as teams split apart. I talked to them about the costs of consensus in decision making. I'm not sure how much it helped.
Prikaži ovu nit -
This is also why we haven't added very many people over the past six months. To Dr. Brooks' point, the cost of consensus goes up when you add new people. They have to be brought up to speed, they cause teams to split apart, etc.
Prikaži ovu nit -
Once things settle back down, beta goes down as well - people are dedicating more of their time to productive measures. Obviously the simple interpretation of the original tweet - that once the team grows you have permanently more communication - is wrong.
Prikaži ovu nit -
But I also see deep wisdom in the observation that the number of interactions goes up with the square of the number of participants. It's a call to be aware and intentional about the increasing costs of consensus so that you don't end up getting less done with more people.
Prikaži ovu nit -
PS. A big thanks to
@vijayravindran, who encouraged me to keep working on this thread when I was ready to give up trying to express myself in ... well ... coherent ... 280-character chunks.Prikaži ovu nit -
PPS. As always, if you like my stories, please also consider following these people and reading their stories. Since the tweet below, I've added
@slooterman@taralconley@DadTrans@mssinenomine@ToriGlass@debcha@graceelavery@Cal__Montgomery et. alhttps://twitter.com/tacertain/status/1132660902963044353 …Prikaži ovu nit -
Due to popular demand, I have bloggified this thread:http://blog.tacertain.com/USL-and-MMM/
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.
But that doesn't magically partition either the software or the knowledge in their heads. Here's where beta really gets you.