Where is Joel
On building software, living a healthy life and the world around us.

On Building Software

I’ve spent the last 15 years building software. Here are my thoughts on how to do it well.

Introduction

Why Software Matters

Effective software teams enable effective companies. There is no scaled company on the planet that doesn’t employ some degree of software engineering, whether it’s interacting with consumers, understanding data or enabling behaviors.

  1. Software teams can create competitive advantage. Any behavior your organization wants to deliver at scale requires software. For example, if you’re advertising at scale on the internet you are likely using facebook business manage and google ads. What if you want your organization to do something at scale a competitor can not do? You definitionally need to write software. If you are able to buy software that enables this behavior, you competitor can too and your organization is not doing anything at scale your competitor can’t.

    EG The Honest Company was started in 2011 and at the time subscription commerce wasn’t something you could buy off the shelf on the internet. Honest build custom subscription commerce software and was able to rapidly dominate a market that lent itself particularly well to replenished orders, diapers. Being the only player in the space with an auto-replenishment feature gave the company a significant competitive advantage over similar companies.

  2. Software is very expensive. A minimal team of 3 engineers, a product designer and a product manager will cost $550,000 per year. Once you have software running a team that size can spend all it’s time maintaining that software, which will not add any additional business value. YouBy committing to hiring a team and paying them to build a piece of software you’ve committed to a $550,000 yearly software subscription.

  3. Software is additive. Good software makes it easier to change. Bad software is harder to change. Both of situations will continually reenforce themselves. Software teams working on easy to change software will find it easy to hit deadlines giving them extra space to worry about software quality. Software teams working on poor quality software will be perpetually scrambling to hit deadlines, think less about quality and their software will get worse and gradually harder to change. The end result of bad software without any intervention will be that the project needs to be scrapped and re-written. The end result of excellent software will be continuous addition of business value until the software lives it’s useful life.

  4. The cost of bad software is invisible. All your scalable behaviors are enabled by software. Bad software scales bad behaviors and good software scales good behaviors. Your business is nothing but a sum of behaviors. What do I mean?

    Imagine a company that sells scrubs. There are several core behaviors that happen, the company orders scrubs, the company markets scrubs and the company reports on it’s performance. The company is a group of people coming together to do these things and if they stopped exhibiting these behaviors the company would cease to exist. The quality of these behaviors entirely drives the outcomes the company is able to achieve.

    Bad software will create bad behaviors. Imagine there is no error checking in the software the company uses to order scrubs. The company will either make more errors in their orders and not have a stable supply chain or they will spend more time double checking all their orders and have less time available for strategic work like maybe lowering the cost of their supply chain. Over an organization the sum of small inefficiencies created by software can easily go unnoticed all while silently choking that organization’s ability to be competitive. Conversly, organizations with high quality software are enabled to have more effective behaviors and are subsequently more likely to outperform.

What this work is

This is a collection of my thoughts on helping to enable effective software engineering teams. Some of it is backed by research, some is from conversations with a myriad of wonderful people I very much respect, some is from experience. If you’ve landed here in some way this is probably a subject that matters to you.

Software Engineering plays a role in almost every modern corporation. This is demonstrated in the fact that there are 4.4 million software engineers in the United States. That’s about 1 in every 30 working people in the country. I’m specifically writing this for those working in Software Engineering at companies that are “Software Enabled Companies”. What do I mean by this?

What is a “Software Company”

I often here engineers flippantly throw around the quote that all modern companies are Software Companies. All modern companies are software enabled companies, but all modern companies are not software companies and the distinction is important.

Software Company

  1. The product you sell is your software
  2. Selling more product has a theoretically 100% gross margin
  3. Your product scales infinitely enabling easy effective monopolies

Software Enabled Company

  1. Literally every other company. (Aside from mom + pop companies that don’t yet need any software)

Let’s examine this for a second.

Google

Google is a company that produces alot of software. Are they a software company? Yes. Why? The product they sell is software that distributes of ads. To distribute another ad, someone else makes the ad, puts it on their platform and then the theoretical incremental cost to distribute the ad is 0, making the theoretical gross margin 100%. To scale their revenue, they don’t need to do anything. They could all sleep in, not go to work, and the product would scale by virtue of customers signing up and creating more ads.

Tesla

Tesla is another company that produces alot of software. Are they a software company? No. What!? They produce the software that powers their cars, right? Right. But someone else could also produce the software that powers their cars and it would take nothing away from Tesla, in fact it might even lower their costs. They make money by selling cars, not software. Each car they sell has a defined margin and they can only scale by building larger factories. If they all went home and slept their car manufacturing capacity would at best stay the same.

Why is the distinction important?

Building good software at a “Software Enabled Company” is a much harder problem than building good software at a Software Company. The near 100% margin at the Software company is able to paper over many mistakes and enables behaviors that would never make sense at a Software Enabled Company simply because it doesn’t make sense to allocate as much of the companies gross revenue to software.

  1. Companies like Microsoft and Amazon can afford dead projects. Dead projects that eat 5% of their gross revenue also eat 5% of their hypothetical margin. Mistakes to be sure, but not life and death mistakes. At a company that say makes products and has a 30% gross margin, a project that eats 5% of their gross revenue would erase over 15% of their gross margin, which might make the difference between making and losing money.
  2. Software Companies can experiment with technology. Something like React Native could only have happened at a software company which means that Software Companies are not constrained by existing toolsets.
  3. Software Companies have much easier access to talent. They can afford to spend more time doing things like blogging which improves their external perception and makes hiring significantly easier.
  4. Software Companies don’t have to worry about obsolescence of their software. Airbnb will not ever be deprecated by Airbnb, it’s the product they sell. If however you are Tesla and you make auto-pilot software, there is probably a company out there looking to make auto-pilot software and sell it to all the car companies. Their economics makes much more sense than yours, they can build once, sell thirty times making the cost of the software on a per customer basis 1/30 Tesla’s cost. Tesla’s auto-pilot engineers should not count on the auto-pilot software product continuing to exist forever.

All of these factors conspire to make building software at Software Enabled Companies require significantly more rigor and discipline than building software at a Software Company. Add to this the fact that the internet is littered with resources from companies like Airbnb advocating a set of behaviors that works for Software Companies. It can be hard to decide on the right tactics and strategies to manage your team and drive your company’s success. For anyone in this situation I hope you find this a useful resource.

On Software Strategy

On Making Decisions

The two activities of an organization

Everything your organization does falls into the category of “Making Decisions” or “Executing on Decisions”. The focus of organizations often strays to improving execution, while failing address decision making. First we’ll examine a few of the negative organizational byproducts of poorly executed decision making, then we’ll look at some of the reasons that organizations fail to do a good job of decision making; finally we’ll talk about what to do to make decision making better.

Why is decision making important?

Your organization is a sum of decisions you’ve made. Probably some of them were good, probably some of them were not. It should be obvious that if we’re able to bias the decisions towards good decisions the organization will be better off. What we don’t realize is the infrastructure around making decisions will not just impact the quality of decisions, but also the happiness of employees and organizational effectiveness throughout the decision making process. Almost every real decision that needs to be made will lead to some sort of employee disagreement, and good behaviors around making the decisions will make those disagreements much easier to handle.

How to know if you have a problem with decision making

Every organization I have ever talked to has an issue some issue with making decisions. Decision making problems in organizations are easy to spot, but are rarely acknowledged because no one looks for them. If any of the following applies to your organization you probably have an issue with decision making that you need to address.

  1. Members of the organization frequently complain about decision making.

    This might sound obvious. People are complaining about decision making, so we have a decision making problem, right? As a leader, you can be completely oblivious to this unless you actually look, because the symptoms are hidden from organizational leaders. The way this manifests is people standing around in the kitchen will say things like

    It’s so stupid that we’re changing the facebook landing pages to client side rendered pages.

    An organization that is good at making decisions will minimize this kind of conversation.

  2. “Decisions” are not followed through on

    Decisions made through broken decision making behavior are much less likely to get executed. They will lack the organizational and team buy-in to make them actually work out.

  3. Members of the organization complain that other’s are “in their lane”

    No matter what you do this will occasionally come up in your teams. This manifests as people saying things like

    I’m the director of data science, and the the database schema for this project is terrible. No one talked to me about it and it should be my call.

    Having “Lanes” should be the first red flag for you. An organization that understands decision making doesn’t need to be nearly as black and white in terms of who owns what. There’s no reason the director of data science mentioned above should have to be the person who decides on the database schema for a project. In fact, this project may not be the director of data science’s main priority and requiring them to make the decision will slow down the project. Rather than making sure that everyone stays in their lane, we solve this problem by following good decision making behavior.

  4. The reason your managers are giving for doing something is “Because so and so said so”

    Any time ay employee is discussing the “why” behind a decision they should understand the reasoning, even if they don’t agree. Saying “We’re doing this because Jeff said we have to do this” is destructive and usually caused by broken decision behaviors.

How to make decisions

  1. Identify that you’re making a decision

    The fact a decision is being made is, sometimes, obvious. For example “Are we going to change from a homegrown software platform onto a turnkey software platform?” or “Are we going to use SQS or RabbitMQ for our queueing system?” Often, however, a decision being made slips through the cracks. For example you marketing team might set up a new facebook campaign that drives to a landing page that hasn’t been used for marketing before. Maybe the lander wasn’t meant to receive marketing traffic so the UTM parameters aren’t set up correctly, or maybe it’s just slow. Deciding to send a campaign to a new lander is the kind of decision that often goes un-noticed but that can cost an organization both in dollars and organizational trust down the line.

  2. Identify a person to make the decision

    For every decision identify someone in the organization to make the decision. Groups are not well suited to making decisions. As a leader, It can be scary to surrender control over meaningful organizational choices. Remember that you’ve hired people because you trust them and they’re good at their job! More importantly, you may not be the best person in the organization to make the choice. Front lines employees usually have more information and by being closer to the actual problem are better able to understand constraints.

    How do you identify the correct decision maker? Ideally you want an employee as close to the problem as possible, but with enough organizational influence that people will buy into a choice they’ve made.

    Imagine a company that manages software architecture. Something like Heroku. The founder/CEO is likely and excellent software architect and probably very good at devops. The organization will have devops problems and need to make decisions on them. Despite their apparent skillset, the organization’s CEO is not the right choice to make decisions on the devops problems, the CEO probably doesn’t spend nearly enough time in the weeds of your company’s tech stack to make a good decision. On some specific decision like how to fix some sort of persistent deploy bug one of your developers with good social influence in the organization would be an appropriate decision maker.

  3. Make sure the person making the decision understands their framework and constraints

    One side-effect of pushing decisions down to the frontline is that the decision maker will naturally have less context. This means that it’s important that the organizational goals and contexts are communicated.

    Let’s talk again about the Software Architecture organization. The CEO is an exceptional architect, but the CEO is not the most appropriate decision maker. It would be unwise to ignore the CEO’s experience. There is a good chance that the company’s CEO has a direction they want the team to move in. For example, maybe the CEO knows that the team will need to remain small based on budget, so they want to keep the surface area of different technologies minimal. If the decision making developer decided to fix the problem by introducing a new technology, then there has not been effective communication of the companies goals and the constraints the decision maker needs to work within.

  4. Have the person making the decision identify and talk with all key stakeholders

    You’ve identified the decision maker and given them a set of constraints. No matter who the decision maker is, they will not have the full picture. Also, key stakeholders need to be involved in decisions to create buy-in. Your decision maker needs to talk to each stakeholder to understand their opinion and the reasons behind the opinion. The decision maker listening to each stakeholder will not only enable them to make a better decision, it will allow them to address any concerns the stakeholders will have if the decision does not go that stakeholder’s way.

  5. Make the decision

    As a leader, you’ve identified a decision-maker, you’ve made sure they understand the constraints, and they’ve talked to all the key stakeholders. Let them make the decision. This is why you hire people you trust who are good at their job. Be thoughtful about it, but unless it’s a disaster of a decision, go with it. Remember, at this point they should know more about the problem than you do.

  6. Communicate with all key stakeholders what the decision is and why the decision was made

    The decision, now made, needs to be communicated out. If you’ve built a diverse organization not everyone will agree with the decision. This results in problems at organizations that have not followed an effective decision making process. If you’ve done everything right up to here, you’ll be fine. Everyone understands who’s decision it was to make and the framework within which the decision was made. Now you need to communicate to each stakeholder what the decision was and as importantly why it was made. Since the decision maker has talked to each stakeholder they should know each stakeholder’s concerns and be able to effectively address the concerns.

  7. Review

    Any decision should get a retrospective that is blameless. Decisions will occasionally be bad or wrong, that doesn’t make the decision-maker an ineffective employee. This is an opportunity to get better as an organization. Don’t miss it by blaming. Do take the time to scrutinize whether stakeholders were all consulted, the framework was passed to the decision maker, and the reasons were effectively communicated out.

How to implement a decision making process

All of this can be implemented without new “enforced processes” and without new stakeholder meetings, decision meetings etc. Rather it should become an organizational habit. For example in step 4 “Identify and talk to all stakeholders” the decision maker could accomplish this without having any large meetings, or passing through and “process gates”. They should understand what’s expected of them and all conversations can be handled 1-1 or on slack.

Implementing this playbook is easy. First, share the playbook. Second, pick a decision and run with the playbook. Third, have a retrospective on how it went. Rinse, lather, repeat. If you and your managers continually decide to push effective decision making it will become an organizational habit.

Introduction