Shipping Video Calling on Facebook

The agony and ecstasy of building the first version of Facebook video calling.

Sixteen years ago this month, I started as the second engineer hired in Facebook Seattle, which itself was Facebook’s first engineering office beyond Palo Alto. There were four desks and thumping techno music 24x7, including weekends. I’d be thrilled to never hear Like A G6 again in my life. Though the other music was great.

At the time, Facebook was 500 engineers in total, small enough for all the senior people in every major product area to know each other. I was 34 years old at the time, and remained the oldest person in the Seattle office for its first six months. This was before Google+, before The Social Network. MySpace was rumored to be coming back. People forget there was a time when Facebook’s future was far from certain.

I had joined Facebook after working twelve years at Microsoft. I was excited to move back into an IC role where I could focus purely on shipping things, at a company where the speed of shipping reigned supreme. I had no idea the company even had career ladder levels; it’d be an entire year before I discovered I was an IC7.

I chose to work as the only full time engineer on video calling back in 2010, but was eventually joined by some true superstars through V1 and V2:

• Thiago Hirai, who later led projects at Snap and OpenAI.

• Denise Noyes, who rose to be a Distinguished Engineer at Meta.

• Vijaye Raji, the current CTO of Applications at OpenAI.

I knew they were amazing even at the time. But there’s a way in which, in hindsight, one is astounded at the company one kept, at the tremendous serendipity of it all.

The Months Leading Up

Having come from Microsoft, which was at the time perhaps 140,000 employees, it was refreshing to be on a tiny, tiny team trying to ship Facebook’s first version of video calling. We had only a fractional PM and designer, and for a while, I was the only full-time developer. I’d soon be, thankfully, joined by Denise, Thiago, and Vijaye.

The company and the team were so small at the time that we did product reviews directly with Mark Zuckerberg. It seemed nothing but natural at the time, and I appreciated how he pushed us to test the extremes. For instance, Mark once suggested we not require people to explicitly answer a call, but instead for calls to immediately be connected. The idea was to simulate aspects of real life (e.g. a coworker can walk right up to you). We sat aghast for a moment at his suggestion, after which Mark revealed he wasn’t really suggesting that approach, but instead just testing the limits to which we would explore the idea of immediacy and in-person interactions.

Video calling was much tricker at the time because WebRTC didn’t have widespread adoption. Chrome and Firefox weren’t even the dominant browsers. So establishing a call wasn’t anywhere near as trivial as it is today. We had negotiated a contract with Skype, prior to it being acquired by Microsoft, so that the underlying video technology would be provided by their client-side binaries, which then had to be installed on people’s computers before calls could be made.

Skype is proof positive that even a terrible product can be sold three times to different buyers (eBay, Silver Lake, Microsoft). I had been a casual user of Skype until that point, but working directly with Skype in order to ship video calling exposed me to the dark underbelly of terrible technology. At the time, I was told that Estonians working in Skype were local superheroes, treated pretty much as Ajay Bhatt entering any arena, complete with wind-swept hair.

But the code itself performed terribly. Skype engineers were incredulous — positively gobsmacked — when I showed them data suggesting call-establishment success rates of 82%. (Video calls take a few back-and-forth messages before a connection can be fully established. At the time, 18% of attempted call establishments through Skype ended in failure.) When I couldn’t convince anyone at Skype that the connection success rate was so bad, I asked them to instead show me their own data about their call establishment success rates. The answer at the time was — and I insist I am not making this up — “We don’t have those numbers.” The logic, as it was explained to me, was that the only truly useful metric was user-reported call satisfaction. Never mind the Skype UI didn’t ever ask you how your call was if it never got connected in the first place. It was a philosophical argument akin to trees falling in forests.

A few individuals I worked with at Skype were stellar. Perhaps the lesson is that even a few stellar folks can be trapped within a subpar organization to produce terrible products. Or perhaps it’s a big-fish/small-pond lesson around being skeptical when people treat you as a rockstar, to look around at the broader world to be sure you’re as amazing as people say you are.

Turning the eye back on ourselves, a big benefit of using Facebook Messenger’s data channel for initial call establishment was confronting the fact of Facebook’s own poor delivery reliability back in 2010. Through video calling’s logs, the team got much stronger evidence around the percentage of messages which simply went undelivered. And, just as the Bluetooth stack on iPhones had poor reliability until AirPods were released, video calling gave Facebook a strong reason to improve underlying infrastructure in order to make a feature possible.

Factoid: the ring tone of Facebook’s first video calling feature was F-A-C-E, a tidbit which those with perfect pitch no doubt noticed upon its release. This stroke of brilliance was the brainchild of an audio designer who I don’t remember meeting in person, but whose cleverness lives on.

Shipping the Product

Facebook was so small at the time that I was one of the key people meeting with the press to launch the feature. It felt a bit unnatural to get “press-trained” by a member of the PR team prior to being allowed to speak with reporters, but I can see why that’s considered a required step at most companies (prior to setting The Crazy Engineers™ loose to say whatever it is they want).

The six weeks after launch were the hardest sustained number of hours I’ve ever worked in my career. I moved our young family down to Palo Alto because I knew work would be crazy. Right after launch, all the bugs started surfacing, requiring hotfix after hotfix. I worked literally all my waking hours, starting around 8am and ending around midnight every day. My wife remembers it as an idyllic time for our toddlers. They visited places like the Palo Alto Junior Museum & Zoo and the local library almost daily.

Fireside True Story™ Time: During the post-ship crunch, I ate nothing but junk food 16 hours a day.

Sun Chips were a favorite; Flaming Hot Cheetos with chopsticks; protein bars; cashews; many bowls of Captain Crunch; chocolate bars; cans and cans of sodas; Chicken McNuggets.

Surprisingly, at the end of six weeks sleeping terribly and eating like a calorie fiend, I had lost more than 10 pounds. Perhaps it was the stress. Or perhaps the brain burns a ton more calories than I had given it credit.

By the time the post-ship crunch was over, the product was in much better shape, connecting more reliably and delivering a new experience many people hadn’t had with their loved ones. Facebook video calling was the first time I had spoken with my parents over video. It felt like magic to bring people closer with a free, easy-to-use form of video calling.

Sixteen years later, after the multi-year crunch mode of COVID, video calling is a staple in people’s lives. WebRTC has made it so that even small startups can release products which include video connectivity. Everybody is used to reminding others to unmute themselves. But there was a time sixteen years ago where none of that was a given, and it was wonderful to experience the baptism of fire that was my induction into the first version of Facebook Video Calling.