Code Review Heuristic: Work You Throw Away

I have been finding it challenging to put my thoughts about code reviews into a linear format. I've realized that it is because code reviews are a part of the broader and context-specific work of engineering a productive software development process.

Rather than talk about code reviews on their own, today I'm going to try something different: I'm going to talk about a specific heuristic you can look at. These heuristics aren't metrics you can optimize, but they can offer insight into the health of your software development process and suggest tools you might try to address them.

The first thing I look for:

How often does the team discard a set of changes and go on to solve the problem a different way because of the discussion in the code review?

The answer should be more often than "never" and less often than "usually". 

If The Answer Is "Usually"

In cases where most approaches to changing the system don't work the first time, the team is either missing norms, skills, communication or an architecture that supports their current work.

Code reviews can be a tool to address all those issues, but if you see lots of PR churn they aren't being enough.

Too Many Silos

If information about a change is surfacing for the first time during code review, is often a signal that there aren't enough conversations happening.

I have never seen this problem occur when the team is pairing or doing ensemble sessions with any regularity, but those techniques aren't the only ways to boost communication. The first thing to try is resolving whatever is keeping people from talking. The technical review process may be too heavy-weight for people to be willing to go through it. Maybe tickets are handed down so tightly-scoped developers are simply executing on pseudo-code, or the team isn't talking through user stories together to identify technical requirements. Potentially a scarcity mindset has taken root and people are concerned about taking the time for casual discussions, or worse being seen to take the time for casual discussions. Cutting scope, adding slack and creating specific spaces to talk can all speed up development by reducing rework.

Whatever the reason, when the only venue for architectural input and technical requirements to come to light is code reviews, it is not surprising that most PRs won't satisfy them purely by luck. The solution is to seek out times and spaces for wide-spread conversations, in whatever formats might work for your team.

Unresolved Disagreements

The other root cause I've seen is a lack of convergence.

If the team doesn't agree on what properties a good architecture has it creates churn. It is also frustrating if different reviewers provide dramatically different feedback: developers give up on finding a good solution and just try to satisfy whoever it is who is keeping them from merging today. If reviewers expect people to write code in a specific way or demand changes without explaining what purpose those changes serve it creates a sense of learned helplessness, where good code and actual agreement is impossible. 

Even when everyone is trying to be collaborative, you can still end up in a situation where there is no common agreement. There may be too many private discussions, or discussions within separate cliques who come to very different conclusions. If a team isn't confident in its ability to engage in productive conflict, people may avoid public discussion specifically to avoid stepping on an undetected third rail.

Regardless of the reason, techniques like event storming, ensemble programming, or question-only reviews can help a team converge on a shared culture of practice. If those would be new to your organization, even just doing a quick five-whys review of whatever was discovered in the code review processes that week at the team meeting can be a good start. It gives the team a facilitated place to practice conflict resolution, and spreads whatever was learned to the whole of the team.

If The Answer Is "Never"

This is worse. When they don't include architectural feedback, reviews aren't able to build quality into your code. There are always things that only become obvious when you take a step back from the painting and look at the bigger picture. If those either aren't being remarked on or aren't being fixed when they are noticed, the code that ships will be worse than it could be.

There are a lot of reasons those opportunities for improvement may be being missed, but I'm going to talk about three here: pressure to merge, a lack of safety, and missing aesthetic feedback.

Pressure To Merge

If developers either don't want to take the time to make code better, or don't feel like they have permission to make the code better, the software will suck.

This is distinct from the situation where there are real reasons for delivering shitty software: in those cases the whole team should be aligned around that vision and the trade-offs made in the particular PR, as well as a plan for how those problems will be addressed once the short-term value opportunity has been realized. If you are going from short-term value opportunity to short-term value opportunity that is still a problem, but those situations do sometimes come up.

But usually the reasons businesses push developers to write shitty software are fake. Invented deadlines, arbitrary OKR review cycles, misuse of velocity metrics, low-trust relationships between product and development, promotion systems that incentivize only short-term feature delivery, and so on and so forth. The list of ways businesses manage to shoot themselves in the foot when it comes to their investment in tech is long and varied, each one making future development more expensive than it needed to be, increasing the fixed cost of the system, and hurting developers' health and well-being.

As a business leader, if you discover that people feel pressured to merge PRs as soon as possible, that is within your sphere of control. Dig in on why, and figure out how you can set up the incentives so developers believe in their soul that you want them to build software in the way that will be most valuable to the business. Enough companies prefer legibility to business value that it's a non-trivial task, but if you get it right the competitive advantage is enormous. 

Hidden Ignorance

Sometimes we are missing a community where it is safe to express ignorance or confusion. If people feel like they have to come to the table with declarative statements and concrete answers, we miss the opportunities to clarify the semantics of our system and create graceful architectures.

There is no one right way to write a piece of software, and a good-enough-to-function way may still be awkward enough that it's a missed opportunity. Building a culture where it is safe enough to ask questions or express opinions even if they are unlikely to change the code review ironically is what opens up the door to discover the long-tail of major changes that have the greatest payoff to the business.

We spend a lot of time and money explaining code, reading code and communicating about code. Any change we want to make requires starting by understanding the system how it is now. We make that work visible and correctly incentivize creating easy-to-explain architectures by having people ask questions in code reviews rather than struggling alone to find the answers. When people believe it is possible to have the software make sense, and feel entitled to architectures that can be explained to them, we end up with software that is reliable, extensible and maintainable.

Unfortunately, this useful work is socially risky.

To build the most business value, we are asking developers to set aside some of our most basic social instincts. There are some tricks that make that easier, like responding to any question with "thank you for asking!" (and meaning it). Engineering leaders can model being ignorant whenever they spot an opportunity. We can also make an effort to respond to questions we don't know the answer to with enthusiasm and curiosity and collaborative learning, so that we aren't putting each other on the spot by asking a question.

It can help to just explain these dynamics. "If you have a question about the code, it is just as useful for the author of the code to know that you had that question as it is for you to get it answered" is one of my go-to phrases. I also like telling a story about how as a first-time manager I had a report with 30 years experience and in his first week he asked me more questions than any intern has ever been able to come up with. I tell that story and challenge new interns to prove me wrong. So far none of them have, and in the process their code reviews have been amazing feedback on where my code can be more accessible and exemplary.

Sometimes those social fears are also well-founded. Facilitated ensemble sessions are particularly helpful as a place to debug any interpersonal dynamics that are contributing to our lack of useful ignorance, as well as opportunities to improve the situation by providing public modeling and corrections.

Missing Aesthetic Feedback

I have a separate post exploring why aesthetic feedback is valuable. Here it is simply enough to say that without taking into account aesthetic feedback, your architecture is unlikely to randomly evolve into something suited to this program's particular context. We are trying to optimize subjective properties of the system: no linear measure can provide more than the roughest of estimates.

There are a couple of common reasons for aesthetic feedback to be missing: programmers might not have learned to articulate their aesthetic reactions, they might fear the social consequences of sharing their aesthetic reactions, or they might have been explicitly instructed to keep their aesthetic reactions to themselves.

They Were Told To

Some time back Microsoft published a series of blog posts confidently declaring incredibly misguided advice about how to provide "useful" code reviews. These blog posts were based on surveys that weren't looking at any of the outcomes we would actually care about, like code quality, learning, value of the software produced, etc. Instead, the stated goal was minimizing time-to-commit of each PR. You could speed that up to infinity by simply eliminating code reviews, but instead they wanted to keep the reviews while discarding most of the value they can provide. Thus their advice was to never say anything positive, subjective or indeed comment on anything other than a specific error in the code.

This advice does successfully minimize conflict. It just also minimizes the usefulness of code reviews. It's basically like economists who write editorials telling you to give your family cash for the holidays: it makes sense only if you completely misunderstand the point and don't believe human relationships have value.

Unsurprisingly, when you forbid people from talking about architecture in code reviews, code reviews don't help you build good architecture. The up-front design process that approach substitutes, in addition to involving significant duplicate work and relying on people guessing right, can't do the whole job. Up-front design decisions are made before the system itself has a chance to provide feedback on how its architecture supports (or doesn't support) a particular change.

This advice has spread far enough that I believe it is usually worth addressing directly. You can't assume that the developers you hire will know all the roles code reviews play in building quality. Make sure your code review philosophy is highlighted when on-boarding new members of the team, and incorporate it explicitly into your leveling guide. Have people in positions of power and leadership state frequently that the goal of code reviews is to cultivate practices that build good software, not to find bugs. And then provide feedback and support to engineers who are treating it as a bug hunt instead.

If you are working with someone who believes in this advice, you can try pointing them to these blog posts. Alternatively, you can arrange to do some synchronous code reviews. This allows you to observe their non-verbal communication as they read the code, and by asking questions of them you can get at the information they were taught to withhold. It also helps build a relationship of trust: when you embody curiosity about their opinion and gratitude when they share it, it tells them that you understand you are all in this software project together.

They Don't Know How

Unlike most professions that involve design, the majority of computer science programs don't incorporate aesthetic critique as something students are taught. Software development management is often focused on what is measurable, like velocity, rather than engineering a context where it is possible to write quality software, and managers often aren't reading enough code on a day to day basis to have their own honed aesthetic sense. Our profession has grown so fast that even in communities where aesthetics are prioritized, we haven't aligned on vocabulary and practices.

The way we get better about talking about aesthetics is to do it, so the solution to this is creating opportunities for people to practice. One technique is to read a book like Implementation Patterns as a group: it highlights the choices we make as programmers and design principles we can use to guide them.

Another fun exercise is reading pieces of open source code and talking about them like they are poetry: what was the author trying to achieve? What techniques did they use to communicate that to the read? How successful were those techniques? How else could they have gone about it, and how would that have changed the code? 

They Don't Want To

The thing about subjective feedback is that it is subjective. It is entirely possible for someone else to disagree with our aesthetic judgement and neither one of us to be wrong.

That is hard for some people to deal with, especially people used to being able to get an A+ by finding the right answer. Ideally in those situations we become curious about what our differences are. What context do I have that you don't? What do you value that I wasn't taking into account? But we have probably all seen situations where the resulting engagement wasn't anything like that productive. One or two bad experiences can scare people off from sharing their judgement.

Leaving positive comments on reviews can help break through that impasse and start building trust. Realizing that positive reactions are just as much feedback on our architecture as squick reactions makes talking about aesthetics less socially risky. Once people are in the habit, they can start sharing times when they notice an opportunity for more symmetry or following up when the semantics of an object aren't clear to them.

It is also useful to note that receiving critique is a skill to be practiced, just like giving it is. If people on your team struggle with being enthusiastic about feedback, consider highlighting the advantages of developing these skills. To progress in technical leadership, being able to write accessible, inclusive code that makes junior developers productive is vital. It is a lot easier to develop that skill if you can incorporate vague feedback delivered by someone new to offering it.

If The Answer Is "Rarely"

It might mean everything is great! Then again, we might be in a mix of dynamics from both failure states at once.

This is why these are heuristics, and not rules or laws or metrics to be measured. There isn't a specific number of PRs I aim for my team to end up completely throwing out. Instead, the most informative signals are qualitative. I keep an eye out for situations where someone would like to rework something but doesn't feel like they can, to see if I can address the blockers and pressures they are feeling. I also keep an eye out for people reworking things purely out of people-pleasing instincts, without personally agreeing with the changes, and encourage deeper engagement on those teams.


Don't give up on stepping back from the painting and taking the whole picture in. 

We can spend all the time in the world trying to guess the right architecture before we start coding, but until bytes meet silicon it is all just a guess. Unless a product is so stable and slowly-evolving that you reliably guess correctly, it is better for both the software and the company to make it cheaper to make architecture changes once you know what changes will be asked of the code, rather than cutting off optionality from the start.

Process can make it impossible to write quality code. We can use tools like code reviews in our pursuit of the goal. But in the end only developers can build reliable, maintainable, valuable software. The company benefits when we advocate for the conditions that let us do our job well.

Software Architecture Is Aesthetic Work

Thinking about why aesthetic feedback on code is so valuable led me to the question of why the feedback techniques we have for procedural code don't work well for Architecture. There have been many attempts to lint architecture decisions, or write unit tests for them, or create frameworks people can plug together like libraries, or create prescriptivist languages, and I have yet to see any of those techniques produce the desired payoff of graceful, cheaply-modified code.

It led me to this realization: architecture is largely semantic work, rather than syntactical.

A well-architected system could potentially end up running the same byte code as a poorly-architected system, and yet be more valuable to the business because it can be more-cheaply understood and modified by people.

Good architecture lets developers make accurate guesses about how the system will behave and how parts of the system we have not read are structured. These programs exhibit the property of a "well-constructed plot", as described by Lee Devin and Robert Austin in their book The Soul of Design. Quality software embodies a coherent set of patterns that work together to fulfill the expectations of the viewer. 

In a system where our expectations are fulfilled, future changes are less likely to have unanticipated consequences, directly preventing bugs. When there is unintended behavior, it is cheaper to debug and correct. The coherent understanding makes it easier to observe the system, and that information can support impactful improvements to system properties like performance, security and resiliency. When we use Domain-Driven Design techniques to align our semantic concepts with our users', we can even use that well-constructed plot to provide a predictable experience to the user.

Unfortunately, we can't have a computer measure that quality. Not for lack of trying; people have come up with a huge variety of heuristics attempting to turn architecture quality into something syntactical we can statically analyze. Reams of papers have been published attempting to flatten the question of suitability into an automatic, linear operation that didn't need to involve humans as subjective participants.

It turns out to be is mathematically infeasible to judge the quality of an architecture in isolation. The properties of coupling and cohesion are only defined relative to a change being made to the system. Referential systems and the semantics of language only carry meaning when read in the cultural and linguistic context a reader brings to the work. Software architecture is inherently postmodern. 

The value of software is created in the relationship between object and subjects across time and space.

Linear models are incapable of capturing the multi-sided, participatory, evolutionary nature of software quality, just like conventional management approaches are incapable of capturing the value of good design. Just like companies that are able to buck conventional management strategies to build extraordinary products can be inexplicably successful, software development teams that are able to buck legibility-obsessed software development conventions to prioritize subjective quality are able to succeed where other approaches fail. 

Luckily there is an easy pattern for collecting distributed, contextual information: ask each node in the system what it thinks. This is why code reviews, collective code ownership and pair programming can be effective tools for quality: they function like MapReduce. Each developer participating in the system contributes information about the quality of our architecture relative to the variety of changes we are actually making.

This is also why having an on-team customer is so much more effective at creating quality than the more-linear approaches of product or project management.

Distributed approaches would still be impractical if each node had to do a lot of work to come up with the answer. Luckily, humans have evolved the capacity for aesthetic pattern matching. Our reactions of pleasure and disgust give us information about whether an object gracefully fulfills the purpose we have in mind. As Devin and Austin explore, it is particularly informative about how cohesive and self-consistent a system is.

Valuable architectures exhibit low coupling and high cohesion. As Kent Beck has described, to accurately judge coupling we have to consider the system in its entirety, whereas cohesion is a local property that is often easy to increase. Similarly, our aesthetic sense is mostly tuned to judge cohesion. The sense of satisfaction we feel in reaction to a coherent, self-referential whole lets us determine at a glance how coherent our code will be after a change.

Rather than struggling to find ways to computationally judge contextual value in isolation, it is more productive to optimize the cyborg system of software+collaborators. We don't need to erase human judgement to build extraordinary software: we need to embrace it.

A Two-Party System Makes The Primaries Important

Student loan forgiveness demonstrates why participating in primary elections is fantastic.
This wasn't Biden's priority and it didn't come from his base. If it hadn't been for the primary, this wouldn't have happened yesterday.
It was one of Warren's priorities. The bill that made student loan debt unable to be discharged in bankruptcy is what got her into politics in the first place, and mitigating that harm was high on her list.
I voted for Warren. That vote became part of the committed 15% support she got across the primaries, even when it was clear she wasn't going to win. When she dropped out and endorsed Biden, it was in exchange for him taking on some of her priorities: student loan forgiveness was one of them.
This means my vote wasn't wasted: people who had less support in the primaries couldn't negotiate for as much. Support in the primary translated into representation in the Democratic platform. Biden needed Warren's 15% to win in the general, and in exchange we got student loan forgiveness and Janet Yellen in Treasury.
Several things went into how effective that was. For one thing, Biden was confident that Warren could deliver her 15%. She hit the road for him with the same enthusiasm she had campaigned for herself, because it wasn't about her: it was about her priorities, and campaigning for Biden was campaigning for her priorities. We voted for him with almost the same enthusiasm we would have voted for her, because it wasn't just about her, it was about her priorities. And we knew that he knew he would need us again, both this fall and two years from now, so we could trust that he would follow through. And he did.
If you ever get frustrated with the "two-party system", I highly recommend getting involved earlier in the process. In American politics, we form the coalitions first, and then we select which of those two coalitions we would like to be the government. Forming the coalition is messier, but it is also a lot more satisfying than showing up at the end & picking between coalitions someone else designed.

In the primaries, it is strategic to vote our hearts even when our candidate won't win, because that is how we get our priorities adopted as party priorities. And then we go knock on doors when the general comes around to make sure our coalition has the power to get them done.

For Delightful Code Reviews, Say Nice Things

A rebellion is brewing.  Ideas like post-commit reviews or even a return to cowboy coding are gaining traction over the unpleasant & unproductive experience that is the bug-hunt code review.

This is unfortunate, because code reviews are one of the delightful parts of our profession. They let us shape and revel in the things we build together. They let us be confident in our work, and demolish imposter syndrome. They are a powerful tool for building livable code with raptor numbers greater than one. While they aren’t the only way to achieve those benefits, unlike ensemble or pair programming they work across time zones and give people extra space.

The problem isn’t that code reviews are bad; it is that they are too often done badly.

Many software developers were introduced to code reviews via impersonal tools or corporate policies that require them. Those unfortunate programmers have never experienced a delightful code review and have no idea how to perform one.

While I can’t give every reader the experience of receiving a delightful code review, I can share with you the tools I use to perform them. Some of those tools require a supportive context or established relationships to work, but there is one that no matter where you work you can start using today:

Say nice things.

As you read the code you are reviewing, pay attention to how it makes you feel. Any time it inspires a a spark of joy, any time you feel yourself smile, leave a comment. 

If you don’t know why you felt joy, that’s okay: your comment can be simply “this delights me”, “:-D” or “Nice!” Your coworker gets to know you appreciate their work, and you get to notice which bits of our work you enjoy.

If you want to take it further, level 2 is figuring out what about that line made you smile. Maybe a name makes sense, or an API is elegant, or you recognize a design pattern used appropriately. By leaving a more-specific compliment, you give your coworker the opportunity to delight you more in the future. 

Level 3 is identifying what doing that good thing accomplished for you as a reader. This not only gives your coworker the chance to delight you; it lets them know the context where doing it again will be similarly helpful. It gives them information they otherwise have no way to learn.

A level 3 positive comment might be something like, “Great job naming this Fire Break! `summonCredentialsFromTheDeep` accurately communicates the monstrosities that lie in those depths. If something goes wrong with credentials, I will definitely know where to look, and it leaves a clear marker that I might want to Tidy First if I need to modify that code.” 

For this to pay off, you can’t fake it: you have to actually figure out what code you like. It is important that you actually enjoy the code you are complimenting. This isn’t some shit sandwich technique: if you don’t have something nice to say, for goodness sake don’t make something up.

It is also important to remember that joy is subjective. It is impossible to be wrong about what you enjoy because it is impossible to be right about what you enjoy. Your joy is your own.

The great things about compliments is that they ask nothing of your coworker. You aren’t trying to get them them to change anything, or telling them they are Wrong[tm]. If they take the critique personally, they have to feel good about themselves. And it is a lot more satisfying to receive that a bland, impersonal “LGTM”.

That doesn’t mean it won’t ever change the code. It may turn out that your coworker wanted to accomplish something different. If how you read it wasn’t what they meant you to read at all, they now have the chance to more accurately communicate their intention! But even then, you still genuinely enjoyed the thing they did. Even if it code ends up changing later, nothing changes your experience of delight.

Compliments are thus a safe way to move code reviews beyond bug hunting. It shows people that aesthetics are relevant to code quality. It establishes that our subjective opinions of our coworkers’ code is a relevant topic, and it establishes that without needing to ask them to do anything to accommodate those preferences. It lets other developers to think about whether they agree with your compliment, and it invites them to leave subjective comments of their own.

But even if no one else got anything out of these comments, I would still leave them. Our trade is fun, and it is worth taking the time to remind myself of that. Not every piece of code we write will gracefully communicate the problem and its solution, but when one does it is a wonder worth celebrating.

Enjoying those moments of grace is my privilege as a programmer.

Coupling & Cohesion: How Musk Would Need To Rearchitect Twitter

Wired magazine published an article about why Musk’s Plan to Reveal The Twitter Algorithm Won’t Solve Anything.

Several of my non-programmer friends were interested in this, and we started chatting. Because the idea itself is obviously shockingly out of left field, I discovered this was a perfect opportunity to explain the properties of Coupling & Cohesion and why they matter.

Coupling & cohesion are defined in terms of a change you want to make to a system. In this case, Elon Musk would like to open source “the algorithm”, which he defines as all the bits of code that “make any changes to people's tweets, if they're emphasized or de-emphasized”.


I want to be clear that nothing here is based on my experience with the Twitter code base. I wouldn't speak to any private information, and my experience was nearly a decade ago. Things have most certainly changed.

The information from current developers in the article is plenty for us to speculate about how coupled and cohesive the system is with regards to this particular change.


Wired magazine reports that there is “no single algorithm that guides the way Twitter decides to elevate or bury content”. Like many high-traffic platforms, Twitter uses a microservice architecture. According to the Wired article, multiple places in the code may promote or hide content, scattered across a multitude of services.

This is an example of low cohesion. We want to change all the code related to promoting or suppressing tweets at once. That code is scattered in many locations & mixed in with totally unrelated code. It isn't cohesive.

In order to understand all the ways a tweet might be promoted or hidden, every piece of the system involved in any of those would need to be open sourced and the reader would need to understand how those components interacted. This makes the change Musk wants expensive & error-prone.

To make the change easier, Twitter would need to rearchitect their system. This would involve moving all the related behavior together in one place. It would also involve separating any behavior in those components that isn’t about promoting or hiding a tweet. A service or a group of services that only handled promoting or hiding tweets would be high cohesion and possible to open-source.


There are two sources of coupling: code that the code being changed relies on, and code that relies on the code being changed. (Anyone know better words to distinguish those two? Let me know, because that is a mouthful.)

Luckily for Twitter, from Wired’s description it sounds like they are mostly dealing with only one of those two kinds of coupling. If not much else depends on which tweets are promoted or hidden, it makes the change a lot easier.

Wired reports that the scattered pieces of code “perform a complex dance atop mountains of data and a multitude of human actions. Results are also tailored to each user based on their personal information and behavior.” That is to say, the code that promotes or hides tweets is highly coupled to many different parts of the current system.

This coupling could prevent Twitter from extracting the behavior into a cohesive unit. Even if the code was centralized, it would still require understanding code that had nothing to do with promoting or hiding tweets in order to understand what is happening. If it is particularly tightly coupled, it might even be impossible to separate without an intermediate step.

Reducing coupling is less straight-forward than increasing cohesion. Twitter would need to consider why those dependencies were needed & what the purpose the data served. They would then turn that understanding into an interface of some kind, with names that reflect that understanding. Twitter’s current data could then be swapped out for some other source of data that satisfied the same purpose. That would let the system be loosely coupled with respect to this change.


I want to be clear here that nothing I describe here is the result of Twitter’s code being “bad”. Twitter’s code is built in a way that would make this particular change hard, but all code makes some kinds of changes hard.

With respect to different changes, Twitter’s system is already highly cohesive and loosely coupled. Twitter grew revenue 16% yoy last quarter. It has recently made obvious strides in reducing harassment & abuse on the platform. All of that involved hundreds of engineers safely evolving running software.

Writing "good" code isn't enough. Even the best code in the universe isn’t loosely coupled and highly cohesive with regards to every possible change.


I don’t think this goal is useful or plausible or something that will happen even if the sale does go through. I’m just using it here as a concrete example of a sweeping change that someone might want to make to an existing system.

Its very absurdity is useful. It clearly demonstrates why You Aren’t Going To Need It is a powerful approach to managing coupling & cohesion.

It is impossible to predict what billionaire in a midlife crisis will get angry your company didn’t let him post Hitler memes in the wake of his breakup & decide to buy your company to make one specific change. Attempting to anticipate that eventuality would have been a massive waste of time & money.

But when something like that happens, regardless of what the system looks like today we can adapt it. And as long as the system solved the previous goals as gracefully as possible, supporting all the existing features plus this one new change will be as easy as we could make it.

Imagine that Twitter had guessed that a billionaire would get mad about an ad they showed him. They might have spent a similar amount of time & effort as this project will take today making the ad targeting logic cohesive & decoupled. The code still wouldn't be any more cohesive or loosely coupled with regards to the change that a billionaire actually wants. It would have cost a bunch of money to do, making all the other work over the years harder, and it still wouldn't make this change any easier.

Attempting to anticipate the future doesn't help us build systems that can adapt to it.

If Musk follows through on wanting to publish all the code that contributes to promoting or hiding a tweet it will cost Twitter a great deal of money. The result is unlikely to be particularly useful, especially when the greatest factor in whether a tweet is "promoted" is whether other human beings hit the “retweet” button.

But by employing these principles, by first increasing cohesion and reducing coupling with regards to the specific change, it would be possible.

Moderating Discussions Over Video

As many colleges move online, I realize I have a somewhat-unique experience: I shameless ripped off the pedagogy from my small liberal arts professors and have spent the last decade+ applying it at distributed tech companies.  I've facilitated video conversations with anywhere from three to fifty participants, both in the course of my work, as part of reading groups for specific texts, on social science topics like "gender and racial bias in tech" and as part of consciousness raising groups to help foment cultural change.

You all have the advantage that the students have already been interacting with one another and with you; that pre-existing trust makes it much easier.  And you are all going to be doing this for the first time, so you can figure it out together. The best advice I can give is saving five minutes at the end to talk about how you all feel the discussion just went, and if it isn't working in the middle of the class, just stop and have a conversation about what isn't working.

The greatest challenge moving to video is that it is easier for people to check out from behind a screen and not have it be obvious.  The advantage is that if they do, it isn't as disruptive.  I always treat any video meeting as opt-in, and then work to make it easy for people to do that opting.

Basic advice:

  • Switch to Speaker View: when someone starts speaking their video will pop up, similar to you glancing at someone who takes a breath
  • Have the other videos in gallery view, so you keep an I on if people are engaged
  • For up to 10 people I don't bother with raised hands.  For more than that some video tools have a built-in mechanisms: for the rest you can use chat.
  • Have one person, ideally not a participant, take collaborative notes so everyone else can pay attention.  If they are in the same room as you (ideal) they can also manage the queue of people who want to speak on a piece of paper so you can glance down and know who is next.
  • Encourage people to take paper notes, so their screens can just be the video.  Similar, have them use an e-reader/tablet or print out any readings.  The other advantage of this is that you will be able to see when they glance down.


  • Have a way for people to submit things they'd like on the agenda before the call starts: that gives you more time to plan. I typically use Slack for this, but wherever this group of people is chatting works.
  • Establish your facilitation plan up-front and communicate it, even if it is the same as last week. Remember, we've just taken out all the turn-taking mechanisms people are used to using for in-person discussions, so replacing them with something explicit helps.
  • Cover the goal of the specific discussion as well, and frame any expectations ("if you haven't read X you are welcome to observe but not participate" is a common one for our architectural reviews, for example.)
  • Since you can't go "around the room", for smaller groups where you want to go round-robin, write down the list of names of participants on the line so you can call in them confidently.  I like using the same list to keep track of people's contributions later.
  • Plan the emotional arc of the conversation: this is the key to keeping people engaged.  
  • As much as possible, prompt for specific kinds of comments, rather than using open-ended questions.  The tendency for confident voices is amplified, and being more specific draws out people who aren't confident responding to general prompts.
  • People aren't getting the usual signals to stop talking, so don't be shy about interrupting. I like interrupting with clarifying questions: it draws people into the conversation, rather than driving them out, without letting them run away with the time.  Facilitation has to be pretty active to be effective.
  • Provide positive feedback on the process of how people are participating, as well as the comments themselves


  • Use the best camera you can find, and try to ensure your internet connection is high-bandwidth
  • Use a microphone (I like or you phone headset if your computer's built-in microphone isn't great
  • Having someone share notes makes the information more accessible, because it is harder to take notes in the video format with a computer in front of you
  • If you have students with hearing loss, CART services can often integrate with video meetings
  • Using techniques like going around the room can help people who are thrown off by the loss of implicit turn-taking
  • If your students' equipment isn't as good, don't be shy about repeating questions

Beyond all of that, know that it isn't as different as it feels at first and it is absolutely possible.

Rails Quirk: A Period in the URL Can Break The Route

When using Rails routing I came across an odd bug: a URL query parameter was breaking the route.  A URL query parameter without a period? Everything works fine. A URL query parameter with a period? 404.

I eventually found the answer in an off-hand comment in a random blog post, and traced it back to the code.  So that next time I remember what is going on, I figured I'd throw the explanation up here. By default, Rails assumes anything after the period represents the format (see the Mapping class defined in rails/actionpack/lib/action_dispatch/routing/mapper.rb). Which if, for example, you are using the format to determine whether a request should be served by a frontend app can then break the route.

To address this, you have two options.  First, you can follow the suggestion I've seen elsewhere and define your own constraint:

get "*path", to: "react_frontend#show", constraints: { path: /.*/ }
But to address the root of the issue, you can also just tell Rails not to look for the format with a regex:
get "*path", to: "react_frontend#show", :format => false

Reflecting on Diversity, Inclusion and My Self-Alienation

Two and a half years ago I joined LTSE, with the goal of changing the incentives companies face to prioritize short-term profits over everything else.  In May, the SEC approved the creation of the Long Term Stock Exchange, making us one of only a handful of venues authorized to list publicly traded companies.

When this milestone happened I discovered that I still have an internalized voice that says, "if you prioritize hiring underrepresented developers, it means you are de-prioritizing success". I found that some part of me holds an insidious belief that places where I felt comfortable couldn't be "the best" companies.  By demanding representation, this voice said, I was asking a sacrifice of the company I was working for. I had gotten as far as believing that sacrifice was justified, even necessary for the sake of justice, but it was still something I was being granted.

That voice is wrong.

We've built an engineering team here that is racially diverse and gender-balanced. We say out loud that we aren't trying to hire "smart" developers: we are hiring skilled developers who believe in practicing their skills in order to improve. We don't believe in a "founder gene": our tools set out to make explicit the implicit knowledge those folks horde, so that more people with valuable ideas can successfully found companies. My experience here is so different than what I had experienced elsewhere.  I no longer fantasize about quitting the industry on a regular basis. I feel like I can recruit without worrying that I am selling harmful snake oil, and I feel empowered to support people the way they want to be supported instead of the way the industries says we should want to be supported. But some part of me distrusts this ease. Part of me still believed that feeling comfortable must mean something is wrong, and that it is unreasonable to want this comfort "at the expense" of the things that "really matter".

That part is also wrong.

I don't believe that our success here vindicates "diverse" teams any more than not succeeding at this ridiculously ambitious mission would mean "diverse" teams are a failure. This is not a magical Utopia, and I still react to things that happen here with the weight of all those other experiences I have had. But this weekend I found myself crying as some deep-seated clinching, this sense that my basic existence was an impediment to success, loosened a little.  It is not unreasonable to want a community or company that takes me into account. We can succeed with a diverse team, where being a feminist is part of the bar, where we expect "D&I" efforts to be effective, where people take parental leave and no one yells and work is expected to be sustainable. It may even turn out that all of those things make it easier, not harder, to do useful, productive, successful work, rather than just being what it takes for me to not quit.

The part that always told me things could be different?  That part was right.

We are going to be hiring a bunch over the next stage of this project. Many of the people reaching out and proactively raising their hands are people who take for granted that every company in the world has a place for them. Some of them will turn out to be great, but my goal in this next phase is to make sure that other people, candidates who wouldn't think to jump in just because the project had some success, feel invited to join as well. I want them to know this company is for them, in a way it is not actually for all these people who get to assume that every company is.

UX Patterns of Emotional Journey

When building transformative experiences for our users, we begin by identifying the emotion that motivates their engagement.  We then imagine how we want them to feel when we have provided for their need.  Finally, we are left to build something that we think can successfully transform the first into the second.

The only way to actually know if such a design works for a sufficient audience to support the product is to experiment and see, but there are some patterns of UX that can suggest things we might want to try.  None of these is a product all on its own: we also have to actually address a need people have in a way that provides some substantive value.  But since we can offer what we think of as value and still not have people walk away feeling better than when they walked in, this is a toolbox we can come back to to ensure that the actual value we provide is also giving people something they want.

There are many more of these possible: I look forward to hearing about the patterns you have discovered! If you are interested in reading more about the use of visuals and interaction in creating experiences, I highly recommend Understanding Comics and Reimagining Comics by Scott McCloud: they are an accessible entry point into the world of visual and interactive impact.  The Design Of Everyday Things and Emotional Design, by Don Norman, are also great starting points, as well as Theater Of The Oppressed, by Augusto Boal and Impro by Keith Johnstone.

Introductory Language Values

I was having a conversation with some people about languages to use to teach programmers.  I am not a teacher and it has been my impression that language choice matters a lot less than pedagogy when creating good programmers comfortable with all of the possible tools of software development.  That said, I still have opinions and here are the things I look for in an intro language:

* Some things should be SUPER easy, hard things should be possible, nothing has to be fast or maintainable
* Fast feedback cycles between writing a thing and seeing if it works, with easily-visible results.  We want to introduce students to the phenomenal magical power of coding, because that will provide intrinsic motivation to keep going when things are hard.
* Transparent: if you dig down, you can figure out why something does what it does (LISP, Smalltalk and JavaScript all have this property)
* Supports encapsulation and recursion: these are the hardest concepts for most students to grasp, so introducing them early and insisting students use them is valuable.
* Good code should be pretty, ugly code should be ugly: I don't care if it's possible to write terrible code, I care if readable code is obviously readable. Students need to be able to start developing an aesthetic sense of code quality right away, but shouldn't have to write clean code in order to get something working.
* No meaningful whitespace or magical numbers of characters: these are often confusing to people not used to working with computers because in other places we use language those things don't matter.
* Good IDE support: most students aren't going to be used to working on the command line, and introducing version control is a more gentle introduction. This helps with explorability and getting students over the hurdle of understanding that they aren't writing prose: they are building a world with it's own internal rules system.
* Easy unit testing: it's hard enough to teach unit testing even at its easiest, but it is incredibly valuable to start with it because it introduces the idea of interfaces and encapsulation in a particularly tangible way. It can also help students learn to evolve software in small, safe steps.
* Publicly-visible well-written code bases: reading code is just as important as writing code when learning.
* An active, supportive, anti-sexist community: I want students to be able to feel like they belong when they go looking on line for information about what they are doing.
* Doesn't try to be clever or optimize for experienced user productivity: ideally, I should be able to tell a story about the language in a single sentence. "Everything is an object" or "everything inside the parentheses gets run together" or "we send objects messages" are ways to bootrstrap a mental model of the language (consider, for example, how Bootstrap explains LISP to teach students algebra:

Note that many of these are different things than I look for in a production language. I want students to make mistakes that help them learn, so protecting them from those mistakes isn't useful or helpful.  They aren't going to be working on large code bases, so libraries, package management and scalability aren't important. No language is perfect on all of these dimensions, but some are definitely better than others.