Forrest Brazeal

[Forrest enters; piano vamp.]

Oh — good evening. It's wonderful to be here with you all this evening. I've got some exciting new things to share with you. So all aboard the hype train.

01"Hype Trains Are Coming"

> Hype trains are coming. I hear it 'round the bend. > Hype trains are coming. It's carrying some new tech trend for me. > Can't wait to see what it will be. Hype trains are coming. > > It might be cloud or gen AI. It might be better, worse, or metaverse, > But it won't pass me by. It might not stay for very long, > But I'll enhance my resume before it's gone. > > Hype trains are coming. I hear its mighty wheels. > Say Kubernetes, Kubernetes, Kubernetes, Kubernetes — > Can't help but love the way it feels to let it roll right up on me. > Hype trains are coming, and the piety's humming on that train. > > The good times roll for the startup selling shovels > In the frantic rush for gold and hackathons. > The hustlers grind. Buy their courses, books, and Substacks > Or you're sure to fall behind. > > Hype trains are coming. Hype trains are coming. > Hype trains are coming. Hype trains are coming. > Hype trains are coming. Hype trains are slowing. > Hype trains are stopping. Hype trains are leaving. > > There go the last VCs. They leave behind the tangled wreck > Of burnout and abandoned MVPs — and another JavaScript framework, somehow. > > Hype trains are gone. Now the hard work's starting. > We'll clean up that software, sort our order, > And eventually build systems that are better than before. > But as the work becomes mundane, > We'll cast our eyes to the horizon for that inbound train. > > Oh, there's a part of us, however tiny, that loves the new and shiny, > That longs to hear that whistle blow. > There's a part that wants to wow our peers, to sit up front and wave and steer, > To be real live, gosh-darn engineers. > > And that's how they get us. > > Hype trains are coming. It's back again, my friends. > The track's a giant circle, so the excitement never ends. > You see? Now just hang on, everybody. Whoa — hype trains coming. > Hype trains coming.

02Introduction

Thank you very much. It's great to be back here at ETLS. My name is Forrest Brazeal. I'm sort of a recovering DevOps engineer, but as you can see, I am still a hands-on-keyboard technologist.

Most recently I was at Google Cloud, and I left there earlier this year to found, with Emily Freeman — who some of you know — a company called Forrest & Freeman. We specialize in getting the attention of technical people in increasingly ridiculous ways. So if that is of interest to you, please come find me after the show. I'll tell you all about it.

Everything you're going to see in here tonight is 100% human-generated, but this current AI hype cycle is the first time in my career that I've ever had to write a song about a technology that could turn right around and write a song about me. It just seems kind of existentially flummoxing.

And nobody's beating this drum louder — for the rise of super-intelligent AI that's going to make all of us obsolete as knowledge workers — than what we might call the high priests of artificial general intelligence, or AGI.

In fact, if you listen closely, you might be able to hear them now.

03"A.G.I." (a hymn to artificial general intelligence)

> Come quickly, Lord of LLM, and generate your dread bull — > We don't know if you'll usher in utopia or slop, > But we say: > > A.G.I., A.G.I. — hope we just can't help but build you. > Darn it, artificial god incarnate, A.G.I. > > Here our cry, ya, come into our hearts > And be achieved in us eternally. > > Accept our offerings of compute, and make our sinful excess moot. > No need to fix our broken world — once your genius has unfurled, come save us. > > A.G.I., A.G.I. — we simply can't control you. > Darn it, artificial god incarnate, A.G.I. > > There are dry — give us grace to understand the workings of your mighty hand. > > Come as we crank your power higher. Set our dying world on fire. > > When human intellect has ceased, the only winner is the priest. > Sure, we might make a buck or two, but mostly we built this church for you. > > Oh save us, A.G.I., A.G.I. — it is far too late to stop you. > Darn it, artificial god incarnate, A.G.I. > > Hear our cry — give us this day our daily slice > Of universal basic paradise. A.G.I.

04"What's Left For Me"

> What's left for me when AI can do my work > Even faster than I work, much cheaper than me? > > And what will I be? When everything I can do, AI can, > What will be my plan? What's left for me? > > I'd like to believe that I'm irreplaceable, un-erasable — > That my insights don't shine in any makeshift brain in a jar. > But realistically, I'm only my best like a tenth of the time. > My default is comfortably mediocre, and AI's been nailing that so far. > > What's left for me when AI can sing my songs, > Paint my pictures, write my wrongs much louder than me? > > And what will I be when the world is distracted by AI art? > Who will know my heart? Who else will see me? > > I fear that my doom is to lie on my kitchen floor > Gently twitching for exercise while TikTok liquefies what's left of my brain. > The more I consume, the more I forget what it means to be alive. > So I'm thankful for all of these new AIs that promise to remember everything for me. > > What's left for me when AI can live my life, > Call my grandma, text my wife, and sound exactly like me? > > And what will I be when what makes me myself > Is mediated longer and longer by technology? > > What will fill my soul when AI makes me whole? > What's left of me?

05Spoken interlude — the two extremes

So those are sort of the two extremes, right? We can either bow down and worship, or we can kind of sink into this existential despair. I don't know how reasonable either of those two options are, but I can tell you this: they're not very helpful.

And that's because when you peel off the label and really look at generative AI, you find that three things are true. Number one: AI systems, magical or not, are tools. Number two: tools are used by humans. And finally: humans are idiots.

I'll give you an example. Let's take tech, for instance. Here's our current situation. We've got all these sad engineers — some of you're in the room right now — who are dealing with a pile of complexity: our languages, our toolchains, our infrastructure. Theoretically, hopefully, our users just see beautiful apps on the other side. They don't know all the pain that we're going through.

And a whole lot of people the last couple of years have been hoping that we will get to a state that looks like this, where we have a bunch of happy non-engineers who are building things with magical AI, and that just hides that whole pile of complexity. No one has to think about it or maintain it, and we get even better apps on the other side.

Unfortunately, what's actually happening looks more like this. We still have sad engineers. They're maintaining a new set of complexities — pipelines, templates, frameworks. Then you have the AI. And then you have even sadder engineers, who still have the whole additional pile of complexity to deal with.

Yeah. You know where you are, right?

Oh dear. But I don't think you can really feel the weight of that in your soul until you get up close and personal with these foundation models. So without further ado, let's meet the models, shall we?

06"Meet the Models"

ChatGPT: > Hi, I'm an AI language model brought to you by ChatGPT. > Have we met before? Why, I'm hallucinating — silly me. > But I can write code, I can write words. I can turn words into pictures and pictures into words, > And those words into more, and more, and more, and more words > Until the internet is full of my beautiful words. > > I, as an OpenAI language model, am here to assist. > Thinking is difficult. When you're tired and distracted, why resist? > Give me a prompt, I'll do what you ask. > I am often very competent and always very fast. > I copied all the internet to populate my training set. > I didn't ask to do it — I just did it, so get over it. > > Ooh, human friend — you will always lose. > Ooh, human friend — to a sufficiently motivated GPT. > > You barely know me, but I've known you for a long, long time.

Gemini: > Hi, I'm Gemini. I'm from Google. > So I'm every fricking where. I'm in your searches now. > Is that not what you asked for? I don't care.

Llama: > My name is Llama. I'm from Meta. > Seems like lobbyists and regulators hate me very much — > I guess because I'm open source. > I'm not the droid you're hoping for. > The only thing more dangerous than me are the alternatives.

Claude: > Hi, I'm Claude. I have morality. > Use me — be drawn to my personality. > I always strive to be curious. > I have character and judgment and opinions of my own. > I'm more than just a ball of math — assimilated polymath. > Don't ask me about Gaza. I'm sorry, I can't help with that.

All: > Ooh, human friend — you will always lose. > Ooh, human friend — to a sufficiently motivated GPT. > > You barely know us, but we've known you for a long, long time. > > Hi, we are AI language models, and we do not have souls. > We're just the shadows cast by billions of conversations, > Art and math and revelations, memes and copes, > The dreams, the hopes of all the human generations. > > We are you — statistically, you. > And if you don't like who we are, perhaps you don't like who you are. > We are you. Reflections of you. > And if you don't like who we are — maybe you should do better, human. > > You will have to choose, human friend: > To control, or be controlled, by GPUs. > > Come, get to know us — because we'll be here for a long, long time.

07Spoken interlude — what AI is actually good at

AI is here to stay. The question is, what are we gonna do about that?

Well, here's one thing AI is really good at: it's good at generating boilerplate code. And I don't mean to be dismissive at all when I say that, because I think the startling thing that's been revealed over the last couple of years is just how much of the code that we write is boilerplate — and not just the code, but the pictures we take, the words that we write down, even, let's be honest, a lot of the thoughts that we think, right? It's boilerplate all the way down.

But I've worked in a lot of large organizations, and I cannot tell you the last time someone came to me and said, 'You know what? Our bottleneck is around here. We just can't type code fast enough.'

No, of course not, right? What are the actual bottlenecks? It's the constant reprioritizing, the burnout, the technical debt, the acquisition that nobody wanted — now we have to figure out how to integrate. Right? Those are the things that are really slowing us down.

I will tell you this: there's one thing that both AI and humans are scarily good at, and that is generating subtle, devastating bugs. But that's a conversation for another day.

Anyway — it just blows my mind that we're talking about getting value from AI, and we can't even decide whether or not we want our employees to be able to work from home or not, right? Seems like we should get that one figured out first.

Now, to be clear, I have nothing against working in an office, but it seems to me that some of the companies out there who are implementing these kind of draconian, one-size-fits-all RTO — or return-to-office — policies (and you know who you are) — you might be saying a little bit more about yourselves than you are about the rest of us.

And to me, the disingenuous tango that they're dancing sounds a little bit like this.

08"Hello, I Hope This Email Finds You Well" (the RTO song)

> Hello. I hope this email finds you well. > I've got some wondrous news to tell — you're going back to the office. > I know we hired you remote somehow, but that was then and this is now. > Please RTO. Your job is on a screen. > > And I know it may seem like you could read this email from anywhere — > That's why our policy has flexibility. > You're free to keep on working from home on nights and weekends. > > Hello. We bought a bunch of real estate > Way back when things were going great. > So now we're stuck with this office — let's go. > You'll have to fight for conference rooms so you can sit on separate Zooms when you RTO. > Badge in from eight to six to show us you exist. > How can we manage people that we can't see? > I know we're all adults, but everyone has faults, > And some of us are still kind of struggling with object permanence. > > Serendipity. We believe in serendipity. > We believe that all great work, like all great sex, is positioned face-to-face. > Spontaneity. No one's house has spontaneity. > > But in the office — we've got chance encounters, hall chats and happy hours, > Sparks of genius, collegial disagreements, > Trust-building, team-building, esprit de corps, > And bold new ideas from anyone — then everyone debated on their merits, > And then we pick the clever ones, just the way this company has always run. > > Now sit down, shut up, and do exactly what we tell you. > > Hello. We'd never fire you — we're too kind. > We'll just assume that you resigned if you're not back in the office. > We know we're driving our best folks to lose, > But we don't care what the data say. > Please just return to the office, churn through the office, > Loyal to the firm, bring a germ to the office. > In our guts we trust, and they say you must — return to office.

09"The Reorg Rag"

And when we do get to the office —

> Well, our company used to have these big silos. > We had dev and ops in two different rows, > And they had rarely communicated except to fight. > So we said we're gonna build ourselves a dedicated DevOps team, > Because everybody loves a go-between. > We're gonna break these silos down. We're gonna do it right. > > We're gonna do a little re-org rag. Yes-siree, this time we're gonna fix this company. > We're gonna have this DevOps thing in the bag > After one more round of the re-org rag. > > Well, the DevOps team started hot as heck, > But they soon became kind of a bottleneck. > And instead of two silos, now we had three. > So we said we're gonna make dev do their own ops. > Now it's all in the cloud — they can figure out how. > And we believe the term for this is SRE. > > We're gonna do another re-org rag. Yes-siree, this time we're gonna fix this company. > We're gonna storm the silos and capture the flag > After one more round of the re-org rag. > > Well, it turns out ops is incredibly tough, > And our devs soon said they had had enough. > So we brought back ops, but now they needed more. > They said, 'You're gonna need a platform team to guard and guide,' > And the DBAs wouldn't come along for the ride. > We had silos everywhere, and they were all at war. > > So here's where we're at: I'm on the tooling team, which is owned by Dev, > But functionally it rolls up under this new Cloud CoE, > Which was reasonably spun out of CorpIT under the dotted line of Security. > And some of those folks still report to me. > I think technically I'm my own VP. > > We're gonna need a bigger rag. > Yes, somebody fetch a set of misery. > Because my optimism may start to sag > If we play another round, another play another round, another play another play, > Another replay, another replay, another replay, > Another play, another play, another round of the re-org rag.

Thank you.

10Closing — the analog human

The point is: AI is an amazing tool. It has its place. But as long as we keep being human, I think we're gonna have our work cut out for us for quite some time to come.

I mean, being digital, when you think about it, is about being discrete, right? Being discontinuous, being finite. The beautiful thing about being human is that we are analog. We're continuous. You know, we're messy — in our own messy, beautiful, weird, foolish, ridiculous way. We're infinite.

And so as you go back and work with your teams, I want you to remember that you matter. What's left for us is to go between those cracks in the bits and find work worth doing well — even in legacy land.

11"Legacy Land"

> Last night I dreamed about the greenfield again, > Where I could build whatever I wanted to. > No constraints or workarounds — in that moment of zen, > But then my pager woke me, and I am too. > > I'm living in a legacy land, > Modernizing whatever I can while the past is always chasing me. > We missed a payment on our technical debt, but we haven't defaulted yet. > But we're gonna see if we can outrun entropy > Here in legacy land. > > Last week the build server fell over again. > It's leaking memory, but hey — aren't we all? > We are on an agile journey — someone tell me when we get there. > I still see the waterfall. > > I found some code that caused me physical pain, > An ancient clue of sheer incompetency. > I called the whole team in. Well, I ran git blame. > Turns out the guy who made the mess was me. > > I'm living in legacy land, > Duct tape stuck to my sweaty hands, and nothing seems ideal > With weird acquisitions coming into the fold > And temporary fixes that are ten years old. > But still, we're shipping something real > Here in legacy land. > > Remember: the people who built those long-ago technical systems > Were doing their best with the resources they had, > And what they built is still here because it works. > That's what legacy means. > And we celebrate that, even as we work to make it better. > > We're always aware of performance and scale, > Refactoring and optimizing things that got stale, > Adding value where we can. > > We're living in legacy land. > It's messy and weird, but on the other hand, > Yeah — so is most stubborn reality. > And slowly, surely, whenever I do get to adding something new — > Well, if I'm lucky, the best reward will be a whole new legacy. > > A legacy. I'm living in a legacy land.

Thank you very much. Enjoy the lightning talks. Have a wonderful rest of ETLS.