jade rubick

Glue your happy brains together with shared state


Operating under a shared reality is a competitive advantage, but most teams share state poorly. Let’s talk about gluing brains together!

Computers and people both have “state”

In computer systems, state is information that is remembered between interactions. A stateful service is one that keeps track of things that have happened before. Databases are where we typically store state, and we try to design our services so they don’t keep track of state themselves, but keep it in a central place.

People, as autonomous beings, require state. The way our thinking works literally requires it. So one of the great challenges of all organizations is distributing a shared understanding of reality: the organizations challenges, goals, and activities.

You don’t copy state, you serialize it

With computer systems, and with people, you can’t just copy state around. I can’t clone my brain, or even a part of my brain, and give it to you for you to understand perfectly what I’m thinking. So in computer systems, we serialize information to put it in a format that can be transmitted between machines.

Human communication is also serialized. We have to put what is in our minds into words or writing to share it with others. This is necessarily imperfect. It is actually impossible for us to express our internal state to each other, so we have to choose what to serialize, and create a facsimile of what we have in our minds.

Human serialization is lossy, too

Our serialization is lossy — people listen poorly, or use pre-existing state and biases as they listen. We are poor at serializing our thoughts to give to other people. The protocols we use for communication are ill-defined, and ambiguous. And just like computer systems, they require a lot of processing power to transfer even simple messages back and forth.

Errors compound the more people you’re dealing with

This is hard enough with two humans. But as we increase the number of people involved in any group of humans that need a common understanding of something, you multiply the number of people involved. Each is deserializing that information in a different way.

What you’ll often find is that people use error-correction algorithms to verify the messages we hear. After a meeting, you’ll talk over what you heard and double check your understanding. And then the two of you will try and figure out what you each thought about what you heard. Consensus algorithms are hard!

One on one conversations can be inadequate for sharing state. Let’s say you have ten people that you’re trying to get on the same page about something. It takes an incredible number of conversations for that group of people to settle on the same shared understanding. This is especially true if the problem or solution is shifting. You’ll often see a pattern where a couple of factions emerge after a meeting where people don’t get on the same page. They each go through numerous smaller meetings afterwards, discussing possible approaches.

In-person teams share state more easily

It is easier for in-person teams to “share state”. When you sit next to someone, you are exposed to the same information, and it’s easier to build a shared understanding. This is a prime advantage of colocated teams, working in the same time zone, on the same project, in the same room.

Often, in-person teams do not share state well. They work on different projects, or don’t sit together. Or don’t communicate well. But what I’m pointing out is that sitting together does allow for an ambient flow of information that helps with shared state.

Distributed teams often struggle with sharing state

Distributed teams require explicit communication to share state. On distributed teams, it feels like things are the same. But it seems to me like there is more variance in how people receive information.

One particularly bad example of this was when I was working at a company preparing an important project for our user conference. We got two months from the release and found out the engineers thought we were delivering “something” at the conference, even though the leadership had been talking about a GA release for months!

When leaders don’t share state, the results are devastating

One of the most common “state” problems on engineering teams is when the engineering manager and product manager aren’t sharing state. They may each have their own version of the project plans. They might each have different ideas of what is being done or what’s important.

Another variation of this problem is when senior leaders are in a different world from the people doing the work on the ground. This is a common issue, because it’s difficult for senior leaders to understand what is happening locally.

How to merge squishy brains together!

There are a few solutions to sharing state.

Be explicit and build trust

One is to check in frequently and have a lot of explicit communication. The more trust you have with someone, the easier it is to share state, because your serialization protocols are better tuned for each other.

Probe for misalignment

Consensus algorithms require error correction. The human version of error correction is asking questions, and listening. That is our protocol for verification.

Use latency judiciously

Latency in communication is a prime factor in how quickly you can come to consensus. If everyone reads a document within a day or two, you have a cycle of 1-2 days. If you don’t perfectly get everyone on the same page within that cycle, it may take you about a week to get everyone on the same page.

Well run meetings can be synchronous, and have a much lower cycle time. Questions can be done in real time. So when taken advantage of, a meeting can be effective to accelerate state building. However, meetings generally come with a catch. If you don’t achieve your objective within that meeting, scheduling the next meeting has high latency. So you have to make sure your meetings truly accomplish their objective.

Use writing to share state

A solution to sharing state is to emphasize writing. When we write something down and collaborate around that shared document, we are all participating more closely in a shared version of reality. While we may deserialize differently, we’re all interacting with the same “database” of information, and the changes are something we all participate in.

Writing is less transitory than verbal communication, and it is more precise. I believe it’s one of the key skills for a successful remote company to master.

I would pay more to make Slack MORE transitory (all messages disappear after a day), because people could rely less on being able to look up previous conversations. Chat isn’t a shared document, because usually the people involved aren’t starting from a shared state. That’s why chat is effective for solving simple problems together, but not complicated ones. I like the pattern of posting links to information in Slack, and having conversations and decisions made around documents. And to reply to things with links to where that information is being maintained.

One of the best patterns I’ve seen for accelerating organizational leadership work is to write down your best thinking of what the problem is and what to do about it, and share it with a group of people. Ask them to make it better. You can go through a couple of rounds of sharing it more broadly, and getting feedback, but you can go through those rounds pretty quickly — a day or two per round. Then just get started! Your plan likely is better than it would have been because you spent the time to write it down. And it’s much better still since your colleagues have improved on it. And everyone’s basically on the same page about the problem and what is to be done about it.

In any case, the important thing is to be aware of state, and how you’re ensuring your messages are being communicated. Think of it like a computer system, and it might help!

Thank you

Thank you to John Hyland for pointing out some oversimplifications in the post, that could be construed as “in person > remote”.

Image by Willi Heidelbach from Pixabay

Jade Rubick

Engineering Leadership Weekly & Frontline Management & Decoding leadership podcast

Point me at your organizational problems. I advise startups and help in a variety of ways.

Comments powered by Talkyard.