Design docs at Google (2020)

You get the behavior you incentivize. In the “early days”, design docs were a tool to agree on a direction and provide context to your coworkers about what you’re doing. Later, as new people joined at an exponential rate, they were told by well meaning managers to write the docs for perf reasons, and things kind of spiraled from there. Google’s culture became a cargo cult of itself.

Some companies I worked at after Google were reluctant to discuss their promotion process in detail, because they saw what happens when people microoptimize for it.

We do "one-pagers" if there's basically one, simple, straight forward way to do something; however, I'm with google on this one.

If you're designing something, and there's only a single solution under consideration, either there's no design, or you're not being thorough. Choices and tradeoffs are what make design

You have a point in that bureaucracy can needlessly get in the way of meaningful change. It frankly always will to some degree. That being said, part of the way you make sure it gets the least amount of in the way is to standardise it, and actually enforce those standards.

From your description you fall into both categories, which means your colleague was correct. Now, your amount of details is very scarce, so it’s extremely likely that I’m getting the situation wrong. But look at it from an enterprise level perspective or change management, if you allow slight deviations here and there for thousands of employees then you’ll very quickly end up with what would be the equivalent of the difference between Italian, Spanish and French. Which all shares the same Latin “standard” and then deviated ever so slightly.

If you think there is a lot of pseudo work involved with managing the bureaucracy of enforcing a design doc standard then you should see just how much pseudo work it saves.

Used to feel this way on other teams. Almost as if you were expected to write them for the sake of writing them (cargo cult engineering).

Now on a team of many OG Googlers (15+ years tenure) and Design Docs only exist when they’re needed (something that touches multiple systems, something that’s obviously complicated with many trade offs, etc) Otherwise it’s just “write the CLS.”

I have a reverse problem at the place I work. When I ask people to write a very high level design doc for a fairly trivial task they go - "There many more way to get this done so writing these is useless. And as the task is trivial an engineer should be able to figure out one of these ways and do it."

Many of these people are external consultants who have been working with the company for 15+yrs. So, there is a fair standard in place already because the tasks are done by the same people. But then they go out of their way trying to create a boogeyman of "what happens if people don't follow the standards?".

The end result is that either design docs don't exist or woefully out of date. Hence the company has to keep hiring these same consultants year after year on hugely inflated costs.

I'm dealing with this right now at another large company. Being asked to write a document for an integration we've done 20+ times because its part of someone else's larger promo project's design.

This is likely due to culture established by the teams working on old, mature products. There, you're going to work with at least 10 people to launch trivial projects. In my case, it's usually 20~30 people with a blast radius of 100~500. You're not going to casually 1:1 with all of them since you and they are all busy. If you don't get a good review from stakeholders, there's a good chance that some angry folks will come for you and make you roll back your launch.

In these contexts, design docs are meant as an asynchronous communication tool for information heavy topics. And this is so asynchronous, you will talk to those people join 10 years later if you product becomes so successful. I've been saved multiple times thanks to some random design doc from 2010 that explains the weird decision that still haunts us. This probably doesn't work very well on lean/small teams or less complex tasks. But engineering culture usually has its own rationale and context, even if it has become a cargo culture.

Another word for Design Doc = "Shelfware"

The mistake was production of a design doc instead of just writing the code. If it's trivial enough that there's nothing to discuss, you generally just change the code. If it's complicated enough, it becomes a negotiation process where at least 5 different people have to be able to put it into their perf if you want to get it done.

This isn't unique to Google. This is encouraged at Uber and Airbnb as well. I would think its a common practice. The idea is to show that you consider alternatives and did due diligence. Following the way its always been done isn't necessarily a good thing. If its a different product area there could be better approaches unique to that area. Its not fake work, it just proves you actually considered other approaches instead of blindly following status quo.

The problem with stuff like this is that it saps the (finite) energy of your team. Not everyone is a tank. High int lose their hit points and they're gone with all their mana.

the compensation structure in big tech makes people lose their minds optimizing for that next promotion/stock grant. It helps big tech retain talented people but the incentives to look "impactful" drive all the wrong behaviors

Couldn’t you just reference a recent doc and say this is mostly the same with these x differences leading to y alterations for z reasons?

The old school way of fixing that problem is to have a standing change order for routine work so you can point to it and say "my problem is just like that" and get the green tick right away. Though the moment you suggests things like change control people get all weird these days.

Then include templates from other areas pointing out the repeated work. You can’t just hand off a couple of lines in a brief to a dev, especially if they are not aware it’s been done before.

I interviewed a couple of folks from Google recently and it was mindblowing how they "work", it's very unfortunate. I feel for some of the people who only worked there their whole lives (e.g: from recent grad within the last 5y), it is unlikely that they actually know how to program.

That must have changed. I worked there from 2006-2014, back then most design docs were useful and followed the basic structure described in the article (minus the system context diagram).

I noticed early on though, that design docs maintained in Google Docs tended to be of lower quality than those maintained in the version control repository. Not sure if that was just a proxy for when they were written or whether the code review process is just more rigorous than editing things in Docs, but the only time I wrote a large (40 page-ish?) design doc I did it using hand written HTML and we worked through the code review system, as was traditional. I also posted it to the central mailing list and web server and got feedback from employee number 3, which was nice. The central location made it easy to find them as they were sorted by category.

Still I don't recall design docs being important enough to matter for a promotion by themselves. Promos were supposed to be about overall impact, not production of specific artifacts. Of course the system was badly flawed, and often yielded surprising decisions in a bad way, but I don't remember ever reading an obviously-optimized-for-perf design doc back then.

If you can find the website that holds the design docs hand written in HTML from the earlier years, try perusing some of those and see if you find them more useful (or would have done when the systems were contemporary). Some of them old ones like SmartASS were full of detailed explanations of the underlying equations and models, which was quite helpful for understanding how it worked, as well as the explanations of why those approaches were chosen. Reading those helped informed my own design work later. I wasn't a director, just an engineer, and they did help.

There are also some Chrome design docs linked to from the chromium.org website that I found helpful in the past for understanding its design.

There is a wide range of types of design docs, and none of them are useful.

I write them mainly as a tool to get visibility with superiors. I try to make them as fancy as possible to to make them think i am "leadership material". I've been feeding them to chatgpt to rewrite them in fancy language.

It sounds like there's no good review process for these; if they lack substance or are too long, they should never be published. But I take it that the people responsible for the people writing these also get the benefits?

I've seen design docs work when you have a relatively high number of junior roles compared to seniors. It forces the junior developers to think through their solutions ahead of time and justify their decisions, and it also allows senior developers to validate those decisions and give feedback asynchronously.

Granted, I've never worked in an org with more than 30-40 engineers, as I've always worked in startups. I'm sure things are different in big tech, but I've had positive experiences with them.

IMO the uses for a technical document (design doc or a shorter doc) is simple. If you get to the point where you can't hold every single detail about a project in your head at once then you should write a doc. Similarly, if it will take a long time (30 minutes at least) to explain to another engineer, you write a doc, to save yourself time.

IDK how you can think that you never need to write a document

Spot on haha.

The promo design doc is directly incentivized by managers: “if you just see a bunch of CLs, it doesn’t tell a story. You need to write it down as a coherent narrative and get some >L5s to comment on it”. “It’s not for me it’s for the committee - they aren’t familiar with your work so you need to explain it”.

But again, they were no engineer, they were a director, I guess design docs make sense for their position

Yeah I think it’s just high potency ammunition in the middle management[1] turf wars. Not even product managers care much about DDs.

[1]: Liberal definition: anyone who uses the word “cross-functional” colloquially, independent of their actual job title.

My experience has been that the second one means, "I need to communicate with my team/TL what I'm doing/how I'm solving this problem." Eventually, when you go for promotion, you take documents in category 2, and add enough context to them that they become category one.

Yes! Although I think you forgot one, the "please just let me start coding" doc.

Red rag to a bull: "this is best practice". Really? There is no better way to do something given more context or experience? A better practice will never be discovered or devised?

Chromium: https://www.chromium.org/developers/design-documents/

Kubernetes: https://github.com/kubernetes/enhancements/tree/master/keps

The closest I have seen are the Oxide computer company RFD documents, many of which are public.

Example: https://rfd.shared.oxide.computer/rfd/0177

Main index: https://rfd.shared.oxide.computer

I guess in the open source side we would talk about Github projects with the greatest READMEs...

Companies keeping their IP a secret? Weird..

Amazon uses writing to bias for action in meetings. Their culture has DRIs (or dictators) who have authority to make the call, with such authority coming with some level of accountability to the decision made. At Amazon the doc is read, parties discuss, and decision made. At Google the doc is rarely entirely read, mostly commented on, and no decision made until all stakeholders align on a commonly agreed direction.

The amazon way is much less toxic, but does come with a need for writer to invest heavily on a high quality doc. Personally, the Amazon way is better but only works due to their peculiar culture.

Devs straight up tell you that they write design docs for the promo committee. That is their goal, everything else is secondary. Given that they never know which design docs they will use in their promo packet they put everything they work on in design docs no matter how small. There is the idea of a one pager design doc, but usually they grow from one page to ... many. It might be a one week project, but a design doc gets written. I have also had to reviewed 20, 30, even 40 page design docs that took clearly a while to make and at any other company would be simply a JIRA ticket. I have had discussions about how many pages is too many and if we should have a limit etc.

Too many have learned (correct or not) that promo committees want to see that only the author wrote a doc and no one else. This further slows down everything and disincentivizes cross learning. I know software engineers who spend a full quarter or more in isolation writing design docs. The majority of the writing in a design doc is the actual design, but the other 99% is the definition of the problem. Too many times I have been in a review and spent a bunch of time improving the definition of the problem which results in the design needing to be scrapped and most of the doc needing to be re-written.

Worst case scenario is when by improving the problem definition a simple solution appears that doesn't require a complex design. The author having invested a lot of time in a complex design (that historically and many committees still look for to promote off of) will push back against anything that is simple. The difficult trick is helping the engineers to improve their problem definition early in the writing process.

I have seen design docs that have no alternatives presented. The design doc was simply a labor intensive way to write down what needed to get done or what someone wanted to do.

This also results in design docs turning into a bug tracking system if you squint. In a hand wavy way everyone is working on their design doc, not bugs, bugs can't get you promoted.

And OP like to wax poetically about how when you join a team you can just go and look at the design doc like that is a simple and obvious thing. They are often not tracked in any central place. For many teams design docs are not owned by the team or the project, but by the individual because you can make sure no one else contributed. This is again for the promo committee sake. Many design docs you can't get access to, not because they are super secret, but just because. And it isn't like teams have say two or three design docs, no there is a mountain of design docs to read. And with around a two year turn over at Google many design docs get lost to time. At any other company this could be like if you joined the team and I said everything you need can be found by reading all the closed bugs or read every commit message in the main branch.

Anywhere else I would have been grabbed after lunch and would have worked on defining the problem for a few hours with the team at the whiteboard. The senior members teaching the younger in real time how to think about these things and iterating quickly. Most of the time this would be written down in the bug tracking system or for bigger stuff in the project's wiki/folder to be owned by all.

All of the above can be improved and I have worked on improving all of them, but culture is a slow thing to change. Design docs are a good idea as a concept, but there are pitfalls and the way they are being used by many (not all!) at Google is not it.

I'm not sure what the difference is, except giving more context than teammates need, or trying to make a problem more complex than it is.

By and large I didn't notice that strategy working.

On the other hand, for context, people wrote large docs on what the team did and does, what the problems are, etc. and those did tend to be long and exaggerated.

That, and to comply with bureaucratic requirements. And commenting on others' docs, to "show leadership."

I long for design docs that are worth more than their cost.

True for every employer i've ever been at. Your career is mostly steered by exposure, rather than reputation or ability. Design docs are very visible to those above you. At yvery company I join, I propose we start writing design docs. It immediately puts me in good standing with management :)

This makes many docs follow the pattern of a more complex design when its completely unnecessary. All just to get more "perf points" from the people that only have time to skim through the docs.

I read many docs that, to me, obviously made up a decision and showcased the thing we already wanted by the time the doc was started, and 2+ silly alternatives (at least one way too simple and an unnecessary overkill) and contrasted them to pick the thing that was reasonable.

There's almost nothing in the greendoc template, I'm not sure it's at all valuable outside of Google.

Why would you? If you feel the process is worthwhile replicating elsewhere, just use your own words for the couple paragraphs of instructions.

It is useful but at the same time different teams IIRC had different design doc structures. I was overwhelmed by go/greendoc when I was an intern and I just opened an empty doc and started writing based on my team's existing design docs... and that worked better. In the end it is just technical writing.

And that is why Google cannot build products at all. I just had a new pixel 7 fail on me yesterday. All of this pretend "we are very smart" behaviours are also a form of a bullshit job. Companies should focus and rate themselves on actual working products.

"I can’t tell you the number of people I’ve met who left Google because it was limiting their career, even when they were deeply comfortable working there."

This is why they pay so high. It's a trap. That, and the apparent "status" of working there (which has mostly worn off now)

You've expressed my frustration with what I experienced there exactly. But I do wish I could have that pay back...

Also re: "agile" I came into "agile" 20ish years ago in the form of eXtreme Programming, and it looked nothing like the cargo cult that is SCRUM or its imitations today. It was, in the end, a set of precepts to put creative power into the hands of developers and let them just get things done without management getting in the way of how -- but in exchange the customer is given the ability to say what things get done and (to some degree) when. Developers do their own estimates. "You ain't gonna need it". No big "upfront design". Refactoring and testing, architecture and design built in as a constant overhead as just standard best practices, not stories or tasks in themselves. Planning meetings are coworkers hashing things out in a room, and "stories" are sticky notes on a whiteboard expressed in minimal, non-technical terms. Standups are literally people in a circle giving a very brief update, in case anybody else is interested, not a ritual to prove you're attending work today, or to show off.

In this system, design is an emergent property of a creative group of experts working together. It doesn't preclude design documents, and it still involves architecture discussions. But it doesn't require an explicit PRD/design-doc process.

How I'd like to work in a shop like that again. I can tell you Google was the polar opposite. Everything took forever.

If a project is sufficiently large and well thought out then architectural changes can be incorporated without too much extra work relative to the whole.

I'm not fully understanding your comment. What do you think is happening or has happened to IMAP at Gmail?

Carl Icahn's the man for that. https://www.bloomberglinea.com/english/i-fired-12-floors-of-...

I think people are really over estimating how much effort goes into most design docs. Most development just happens and occasionally you slam out a doc to make it easier to justify CLs.

Maybe one out of ten do I see someone who went way overboard but it is really not a time sink for the average SWE.

Which aspect of the Amazon design doc culture that you loved the most?

Design docs are vehicles for discussion. The idea is that they are the most efficient way to communicate your intention, motivation, why not alternatives.

They are one mechanism that fits in a broader working culture. If you’re working solo, it’s an extravagant exercise. If you’re in a huge team, it’ll enable leveraging more expertise across the team and act as documentation.

There are a few different failure modes.

* Valuing output not outcomes is classic misalignment. (writing 40 page docs for promo — this doesn’t work except for very junior folks who are proving the ability to string words together more than deep eng)

* As mentioned, it’s a bit much of solo teams. Other small teams can comfortably communicate through issues (Jira) plus breakout sessions to hash out ideas.

* Engineers also need to be onboarded to effective design doc writing. (note a top commenter frustrated that their first effort isn’t immediately met with adulation, a possible flag.)

Writing about code is hard. Usually this is an exercise celebrated on HN. If you’re in a team, beware if you find your work is always that which doesn’t need to be explained and deeply considered in a shareable doc.

I don't see how design docs are the problem in your scenario

Perhaps read the section dedicated to this point: https://www.industrialempathy.com/posts/design-docs-at-googl...

Some cross-cutting concerns are about more than just code, but things like "how will you monitor this for reliability?" or "this system reads user-sensitive data, how will those reads be audited?" and so on. Usually the answer is short: "use standard system xyz" - but having it there and ensuring it's followed through on are important.

Same here. I found the biggest beneficiary of a design doc was my own thinking process (Xoogler here)

You always want to work in a product that maintains a decent one.

That'd be lovely indeed. I have never seen one that I felt comfortable with, though. And I've seen a few.

When LLMs rose in popularity, I had hopes of having a way to tackle the problem of formulating requirements in a way that would be less painful than it is now.

To me, computer programming is a form of communication, where the requirements are close to humans, and program code is nearer to the machine. On the computer side, we have linters, compilers, and a deterministic system that executes the code. On the human side we have people stating requirements such as "The system should be easy to use."