Five Lessons From My First 18 Months as a Dev

I recently moved from Sydney to London to start a dream job with Twitter as a software engineer. A software engineer! Who would have thought.

Having started my career as a journalist, the title ‘engineer’ is very strange to me. The notion of writing in first person is also very strange. Journalists are taught to be objective, invisible, to keep yourself out of the story. And here I am writing about myself on a public platform. Cringe.

Since I started learning to code I’ve often felt compelled to write about my experience. I want to share my excitement and struggles with the world! But as a junior I’ve been held back by thoughts like ‘whatever you have to say won’t be technical enough’, ‘any time spent writing a blog would be better spent writing code’, ‘blogging is narcissistic’, etc. 

Well, I’ve been told that your thirties are the years where you stop caring so much about what other people think. And I’m almost 30. So here goes!

These are five key lessons from my first year and a half in tech:

Deployments should delight, not dread

Lesson #1: Making your deployment process as simple as possible is worth the investment.

In my first dev job, I dreaded deployments. We would deploy every Sunday night at 8pm. Preparation would begin the Friday before. A nominated deployment manager would spend half a day tagging master, generating scripts, writing documentation and raising JIRAs. The only fun part was choosing a train gif to post in HipChat: ‘All aboard! The deployment train leaves in 3, 2, 1…”

When Sunday night came around, at least one person from every squad would need to be online to conduct smoke tests. Most times, the deployments would succeed. Other times they would fail. Regardless, deployments ate into people’s weekend time — and they were intense. Devs would rush to have their code approved before the Friday cutoff. Deployment managers who were new to the process would fear making a mistake. 

The team knew deployments were a problem. They were constantly striving to improve them. And what I’ve learnt from Twitter is that when they do, their lives will be bliss.

TweetDeck’s deployment process fills me with joy and delight. It’s quick, easy and stress free. In fact, it’s so easy I deployed code on my first day in the job! Anyone can deploy, at any time of day, with a single command. Rollbacks are just as simple. There’s no rush to make the deployment train. No manual preparation. No fuss. Value — whether in the form of big new features, simple UI improvements or even production bug fixes — can be shipped in an instant. The team assures me the process wasn’t always like this. They invested lots of time in making their deployments better. And it’s clearly paid off.

Code reviews need love, time and acceptance

Lesson #2: Code reviews are a three-way gift. Every time I review someone else’s code, I help them, the team and myself.

Code reviews were another pain point in my previous job. And to be honest, I was part of the problem. I would raise code reviews that were far too big. They would take days, sometimes weeks, to get merged. One of my reviews had 96 comments! I would rarely review other people’s code because I felt too junior, like my review didn’t carry any weight. 

The review process itself was also tiring, and was often raised in retrospectives as being slow. In order for code to be merged it needed to have ticks of approval from two developers and a third tick from a peer tester. It was the responsibility of the author to assign the reviewers and tester. It was felt that if it was left to team members to assign themselves to reviews, the “someone else will do it” mentality would kick in, and nothing would get done.

At TweetDeck, no-one is specifically assigned to reviews. Instead, when a review is raised, the entire team is notified. Without fail, someone will jump on it. Reviews are seen as blocking. They’re seen to be equally, if not more important, than your own work. I haven’t seen a review sit for longer than a few hours without comments. 

We also don’t work on branches. We push single commits for review, which are then merged to master. This forces the team to work in small, incremental changes. If a review is too big, or if it’s going to take up more than an hour of someone’s time, it will be sent back.

What I’ve learnt so far at Twitter is that code reviews must be small. They must take priority. And they must be a team effort. Being a new starter is no “get out of jail free card”. In fact, it’s even more of a reason to be reviewing code. Reviews are a great way to learn, get across the product and see different programming styles. If you’re like me, and find code reviews daunting, ask to pair with a senior until you feel more confident. I recently paired with my mentor at Twitter and found it really helpful.

Get friendly with feature flagging

Lesson #3: Feature flagging gives you complete control over how you build and release a project.

Say you’re implementing a new feature. It’s going to take a few weeks to complete. You’ll complete the feature in small, incremental changes. At what point do these changes get merged to master? At what point do they get deployed? Do you start at the back end and finish with the UI, so the user won’t see the changes until they’re ready? With feature flagging — it doesn’t matter. In fact, with feature flagging, by the time you are ready to release your feature, it’s already deployed, sitting happily in master with the rest of your codebase. 

A feature flag is a boolean value that gets wrapped around the code relating to the thing you’re working on. The code will only be executed if the value is true.

if (TD.decider.get(‘new_feature’)) {
    //code for new feature goes here
}

In my first dev job, I deployed a navigation link to the feature I’d been working on, making it visible in the product, even though the feature wasn’t ready. “Why didn’t you use a feature flag?” a senior dev asked me. An honest response would have been: “Because they’re confusing to implement and I don’t understand the benefits of using them.” The fix had to wait until the next deployment.

The best thing about feature flagging at TweetDeck is that there is no need to deploy to turn on or off a feature. We set the status of the feature via an interface called Deckcider, and the code makes regular API requests to get the status. 

At TweetDeck we are also able to roll our features out progressively. The first rollout might be to a staging environment. Then to employees only. Then to 10 per cent of users, 20 per cent, 30 per cent, and so on. A gradual rollout allows you to monitor for bugs and unexpected behaviour, before releasing the feature to the entire user base.

Sometimes a piece of work requires changes to existing business logic. So the code might look more like this:

if (TD.decider.get(‘change_to_existing_feature’)) {
    //new logic goes here
} else {
    //old logic goes here
}

This seems messy, right? Riddling your code with if else statements to determine which path of logic should be executed, or which version of the UI should be displayed. But at Twitter, this is embraced. You can always clean up the code once a feature is turned on. This isn’t essential, though. At least not in the early days. When a cheeky bug is discovered, having the flag in place allows the feature to be very quickly turned off again.

Let data and experimentation drive development

Lesson #4: Use data to determine the direction of your product and measure its success.

The first company I worked for placed a huge amount of emphasis on data-driven decision making. If we had an idea, or if we wanted to make a change, we were encouraged to “bring data” to show why it was necessary. “Without data, you’re just another person with an opinion,” the chief data scientist would say. This attitude helped to ensure we were building the right things for our customers. Instead of just plucking a new feature out of thin air, it was chosen based on data that reflected its need.

But how do you design that feature? How do you know that the design you choose will have the desired impact? That’s where experiments come into play. 

At TweetDeck we make UI changes that we hope will delight our users. But the assumptions we make about our users are often wrong. Our front-end team recently sat in a room and tried to guess which UIs from A/B tests had produced better results. Half the room guessed incorrectly every time.

We can’t assume a change we want to make will have the impact we expect. So we run an experiment. Here’s how it works. Users are placed into buckets. One bucket of users will have access to the new feature, the other won’t. We hypothesise that the bucket exposed to the new feature will have better results. The beauty of running an experiment is that we’ll know for sure. Instead of blindly releasing the feature to all users without knowing its impact, once the experiment has run its course, we’ll have the data to make decisions accordingly.

Hire the developer, not the degree

Lesson #5: Testing candidates on real world problems will allow applicants from all backgrounds to shine.

Surely, a company like Twitter would give their applicants insanely difficult code tests, and the toughest technical questions, that only the cleverest CS graduates could pass, I told myself when applying for the job. Lucky for me, this wasn’t the case. The process was insanely difficult—don’t get me wrong—but the team at TweetDeck gave me real world problems to solve.

The first code test involved bug fixes, performance and testing. The second involved DOM traversal and manipulation. Instead of being put on the spot in a room with a whiteboard and pen I was given a task, access to the internet, and time to work on it. Similarly, in my technical interviews, I was asked to pair program on real world problems that I was likely to face on the job.

In one of my phone screenings I was told Twitter wanted to increase diversity in its teams. Not just gender diversity, but also diversity of experience and background. Six months later, with a bunch of new hires, team lead Tom Ashworth says TweetDeck has the most diverse team it’s ever had. “We designed an interview process that gave us a way to simulate the actual job,” he said. “It’s not about testing whether you learnt an algorithm in school.”

Is this lowering the bar? No. The bar is whether a candidate has the ability to solve problems they are likely to face on the job. I recently spoke to a longstanding Atlassian engineer who said they hadn’t seen an algorithm in their seven years at the company.

These days, only about 50 per cent of developers have computer science degrees. The majority of developers are self taught, learn on the job or via online courses. If you want to increase diversity in your engineering team, ensure your interview process isn’t excluding these people.