Left Stripe, joined an AI start-up

A month ago, I left my job of 5 years at Stripe to join an AI start-up. Here’s some scattered reflections, one month in. No “large tech company vs. startup” philosophical takes, I’m afraid, just some light observations contrasting the day to day.

Some things I miss

Change is hard. There are things I miss from my old job.

1. All the sentimental stuff, my old team, etc. This goes without saying.
2. Being able to draw from a vast catalogue of Slack reacjis. There are rare moments when you really need to reach for :saturn-devouring-his-son: or :i-volunteer-as-tribute:. At my old job, I could reach for these and they would reliably exist. At my new job, I must create them to use them, by which time the moment has likely passed. I didn’t realize how important readily-at-hand reacjis with moderately obscure titles were to my self-expression, but here we are. I’m going to have to up my emoji-creation game.
3. Go links - At my old job there were “go links”, you could type go/name-of-thing into your browser address bar and almost always it would take you to the page you want. At the new place, I spend effort curating a bookmarks toolbar.
4. Jira - Ha ha, just kidding.

AI has playfulness built in

As you might gather from the tasteful fonts and Javascript animations on the front page of my blog, I value playfulness – or at least, playfulness is part of my brand.

Fintech is BYOP, “bring your own playfulness”. You can find ways to be playful if you choose: one thing I liked to do was, in tests or examples, frame them with slightly absurd business models: What if Cupid, Roman god of love, needed to monetize? There is nothing playful about a payments product itself, however. I would certainly think twice before adopting a payments product advertised as “playful”.

AI products, though, are and should be playful at times. This is especially true of emotionally intelligent AI models, like I work with at the new job at Hume. Perhaps the novelty of this will wear off eventually, but in the meantime this adds quite a bit of joy to many of the workaday tasks of software life.

Troubleshooting bugs, for instance, can be hilarious (local bugs on a branch, that is — I don’t really find bugs in production funny). It’s also just fun to prompt the AI with silly things to see what comes back. (Outside work hours, I do this to my two-year-old too — he’s usually a lot funnier than the AI though.)

Phoebe Syndrome

There’s a character Phoebe from the “Magic School Bus” franchise whose catchphrase is “at my old school…” When you switch jobs — especially from a successful, name-brand company — you have to learn to navigate the temptation to say “at my old job…” in technical discussions.

Sometimes, it’s appropriate. If I have detailed knowledge of a decision that worked successfully in a similar context, that is valuable to contribute. I’ve realized, though, in some cases what I have doesn’t meet the bar of detailed knowledge, it’s more like vague knowledge that something at the old job worked a certain way, plus a strong but vague intuitive sense that this way is correct. Even when the vague intuitions are right, holding them actually interferes with examining or arguing for ideas on their merits.

So far I’ve done a good job biting my tongue before “at my old job…” can get out— but I am gaining a new level of respect for my colleagues at the old job who were transplants from big, successful places, who managed to keep their own Phoebe syndromes in check so well.

Two wonderful jobs

Overall, I feel very lucky to be where I’m at and to have been where I’ve been.

My old team is hiring. Stripe — and the SDK team in particular — is an incredibly special place to work. If you like working with a lot of programming languages, getting into the nitty-gritty troubleshooting user issues, but also helping with API design and working with abstract representations of code, I doubt there’s any better team in the industry.

My new company, Hume, is building amazing technology. For the first time in history, you can use the computer with your voice in a conversational way that doesn’t feel like talking to an audio command line. You can interrupt it, joke with it; it can read and respond appropriately to your tone.

For most of the computing age, software has been training us to change our habits to be more compatible with it. The things that are happening now in voice AI can reverse this, if we get it right. Command lines and widgets have taken us a long way, but software will be more pleasant, more accessible, more humane, and more playful when, in addition to controlling it via mouse and keyboard and finger swipes, we can talk to it and explain what we need to do in our own words. It is a gift to be part of this.

If you’re in software – even if you’re some big tech backend observability performance engineer, who won’t be in a position to be building with voice any time soon, I think it’s important for you to have the experience of building with conversational voice. Start by trying the Hume demo (or OpenAI’s advanced voice mode, which has similar capabilities). Then build something yourself: make an account, follow the quickstart and play around a little. (After hello world what I did next was writing voice tic-tac-toe, but that did take most of an afternoon.)

For me, experiencing this technology was compelling enough to cause me to reorient my whole career. You might not have the same reaction, but at the very least, it’s an experience you should have. It is fun, and will feed your imagination.