Twitter can improve software quality by losing most of its engineers

Twitter has fired boatloads of software engineers and more are jumping ship. Most people predict that losing all those engineers will lead to software disaster. But then, most people don’t know much about software and don’t realize what a disaster Twitter software has been for years. With some intelligent leadership, Twitter's software could dramatically improve while dispensing with 90% of their engineers.

Twitter Customer Losses

First, let's note that the situation has users really worried. Customers are bailing the ship:

Here's some up-to-date analysis:

While some say the massive customer losses are due to the new CEO, others are sure that the loss of so many excellent Twitter software engineers will seal the deal.

Twitter Software Quality

Those cool internet software companies – their top people wear hoodies or whatever they feel like. They must be fabulous programmers, right?

Nope. Facebook, for example, produces amazingly bad software. See this and the included links for details on how bad it is – and how they hide as much as possible from their users.

Facebook, Twitter and the rest aren’t better than “normal” software companies, except in how rich they made their founders. Internet software is a horror show, as I document here.

I dove into Twitter in particular years ago to see just how bad it was. I found and documented inexcusable failures, which they went to great lengths to disguise.

I tested doing searches for tweets with Twitter’s own search engine, and documented random patterns of tweets being dropped out or included from searches just days apart. See this and this.

What this means, among other things, that the basis of “trending on Twitter” is based on bad Twitter data – the kind produced by those wonderful, oh-so-cool Twitter engineers.

Twitter hides the awfulness of its terrible software from you

Note that the errors I documented were NOT associated with an error message. They were cases where the right answer (for example search results) could be determined and compared to what Twitter provided. That’s what I did, and found that Twitter blithely would state “here’s your answer,” and then provide an answer that was demonstrably false. It would have been more honest had they given an error message instead. But no -- that would have been honest; Twitter engineers, while terrible at building software that actually works, have become masters of the simpler, nefarious job of masking bad software.

Even with such errors, Twitter would still admit failure from time to time, showing this image:

Since then Twitter has gotten better at hiding their errors. Following the pattern of bad search results, they give you your feed of tweets and you merrily scroll down, reading, retweeting, liking, and whatever. They have learned to almost never admit failure. The fail whale shown above is history. How do you know if you’re getting all the tweets from the people you follow, and in the right order? You don’t! In fact, there is loads of anecdotal evidence that part of how their engineers spend their time is figuring out how to manipulate your feed for a variety of reasons -- including masking the results of their inability to do it the right way -- instead of just giving you a common-sense, complete ordered feed.

Fail whale? I haven’t seen it in years. They can have dozens of servers crash and burn and no one will know the difference. What Twitter does is just pull together your search results or feed based on whatever servers are still limping along with whatever the pathetic Twitter software will give them, and give you whatever crap it has. How will you ever know it’s incomplete, missing important things, etc.? Unless you’re an obsessive crazy person like me and run tests, you’ll never know.

So what do all those thousands of software engineers do all day long except attend meetings while relaxing on comfy bean bags and digesting all the free food the company provides? Obviously not much in the way of useful programming.

The alternative

Small teams of smart, motivated programmers typically out-perform teams of hundreds and thousands of people with titles of “software” employed in big bureaucracies – including in “cool” companies. The concept is simple: most programming organizations are like people who build bridges in times of peace. They take thousands of people years to build. When you build bridges in war, like the bridge over the Rhine in World War 2, you have to build it in a DAY, while under enemy fire. And it has to work. Small, under-resourced groups of programmers operate in wartime mode, getting things DONE while groups of thousands continue plodding away at requirements planning. See this. For more, see this.

I see this all the time in my work of evaluating small, innovative start-ups. The small companies don’t have enough time or money to do things the “right” way. They have to get things done. Fast! The ones that do succeed.

A great example from another field is the author of the first important dictionary of the English language, Samuel Johnson. He produced excellent work, arguably 200 times more productively than the big committee trying to do the same for the French language. See this for the story.

What Twitter does isn’t hard! And what it does hasn’t fundamentally changed for at least a decade! Those thousands of engineers couldn’t make it work well nine years ago when I ran tests, and all they’ve managed since then is do a more polished job of hiding the errors and bad answers.

Whether Elon Musk or someone like him is in charge, the best thing for Twitter and its customers would be for most of the programmers to be shown the door, for the remainder to get with the program of making things actually work and work well, or join their former colleagues as ex-employees. If they really get it going, some programmers who are actually good and, get ready, WANT to write code, GOOD code, may seriously consider joining a re-invented Twitter engineering group.

It might actually happen, since Mr. Musk is violating the most sacrosanct rule of management: getting involved with the people on the front lines of actually doing the work! There's even evidence that he seems proud of his disgusting behavior.

Who does he think he is??!!

Comments

[Lukas Oberhuber]

As always, very thoughtful. But maybe our beliefs about good software are wrong. I’m talking about the belief that software can be: well written; not fail; and be easy to change. Maybe we are too utopian about what is possible?

In my view, software is like a city. Once established, you can’t just rip up the roads and buildings and start over. Sure, you can build a city from scratch pretty quickly (see Qatar for the World Cup 2022), but changing an existing city is another matter. So too for software. Once the software is written, you can wish for the easy yards of simple requirements and rapid development. But you have a city to work around. And just like a city, this is only a problem because the software has been successful. The unsuccessful stuff is very easy to change: just throw it away.