2024-12-23 08:00:00
A friend was telling me the other day about why Google snapped up Character AI. Apparently, the big deal wasn’t just the cool AI girlfriends everyone were chatting to, but that they’d cracked the code on running these things without costing a fortune. Like, doing the AI magic – the “inference” part – at a scale that wouldn’t bankrupt you.
The adoption curve of any new tech usually starts with porn and parasocial relationships. Look at any list of GenAI products with the highest MAU - most of them are about curing loneliness. AI girlfriends, AI companions, AI therapists. But running these services is expensive. Like really expensive. The inference costs can kill you before you even figure out product-market fit. This is what made Character AI special. They could serve millions of users without bleeding money on each conversation. And Google probably saw this and went “hey, this is exactly what we need for Gemini”. Now that massive amount of money they paid for Noam Shazeer has started to look like a pretty good investment.
Think about it. Google is now giving away these flash models for free on AI Studio. You can’t do that unless you’ve figured out how to make inference really cheap. Cheap enough that you can offer it as a free product without destroying your unit economics.
Pre-training has hit diminishing returns. Everyone knows this. That’s why you’re seeing this shift towards chain-of-thought prompting and inference-time optimizations. It’s probably why Anthropic is working on improving Sonnet instead of launching something bigger. And OpenAI is launching O3 and not GPT 5.
2024-12-22 08:00:00
Follow these people on x dot com: samdape, luusssso, FonsMans, mariapicasso, hanslorei, laurentdelrey, realvjy, marty, rjonesy.
Pages: spottedinprod, viewportui, makerstations, LogoArchive, 60fpsdesign, eightbitstories, 4AAAAart.
Places to discover new art and designs: arena, cosmos, mobbin.
Take photos. Create on Midjourney. Read books. Try to copy and then remix things you find interesting.
2024-12-21 08:00:00
I was having this interesting debate with a friend recently about whether product folks need to go deep into the AI weeds. His argument was pretty straightforward: the fundamentals of product management - understanding users, driving activation, figuring out monetization, running experiments - remain the same whether you’re building web2, web3, or AI products. Why not just focus on what we’re good at?
I see it differently. I could make better product trade-offs (scope, impact, effort) because I was once a developer. I’ll admit it, I used to totally gaslight people about how long my dev work would take.
Since then I have worked in product for more than a decade.
So I’ve been on both sides.
Here’s the thing that’s been bugging me lately: I had this conversation about an interaction on our Android app with a developer. He started talking about something called RecyclerView and why what we were discussing would take ages on Android because of some reason. I was not even sure if what they were saying made sense or if they were just throwing technical jargon at me knowing I wouldn’t catch their bluff. My Android development experience was from a decade ago - back when people used Eclipse, and we only worried about Activities and Fragments. RecyclerView? No clue what that does.
Now, do I need to know everything? Nah, probably not. But I gotta know enough to, you know, smell BS when I see it. To figure things out for myself.
Whether you’re founding an AI company or working as a product person at an AI startup, you can’t skip having some baseline knowledge of the domain.
This is why if you look at my Twitter, you’ll see me sharing random AI stuff. There’s no structure, no clear project, just disconnected ideas floating around. And that’s fine. I’m in my exploration phase. I’m okay looking dumb because what matters is whether I understand things better over time or not. Do I need this as a senior product person who doesn’t do much IC work? Probably not. But learning and doing new things is what keeps life interesting.
It’s like when I didn’t know you needed to use Frames instead of Groups in Figma. Now I can create simple design systems too. It’s never too late to learn new things.
2024-12-20 08:00:00
I was talking to a friend yesterday about AHA moments in products. He is working on the strategy for a late stage unicorn startup. His management feels they have a solid enough product already, they just needed to package it better, explain it well, help users reach that magical AHA moment faster.
My point to him was: not every product has an AHA moment.
Remember when everyone in product was obsessed with finding their North Star? Facebook made it seem so simple - get users to add X friends in the first Y days, and boom, they have enough content to stay engaged. Their growth team spent years decreasing the time to add X friends. Cut all friction. Reach the AHA money earlier.
But that was Facebook. Connecting to your friends and sharing your life updates was novel.
What if you’re building the 10th fintech app?
Unless you can genuinely show users why this is revolutionary - make them go “wow, I never realized I could do this” or “I always needed this” - you probably don’t have an AHA moment.
What are some companies that nailed the AHA moment?
Zepto. We’ve had on-demand delivery forever. Swiggy and others were doing 40-minute delivery. So what made Zepto different? A VC friend explained it perfectly - it changed his behavior. Earlier, if his friends came over, he’d stress about getting drinks or snacks. The thought of waiting 40-50 minutes for a diet coke or going down to buy it yourself. But with Zepto? It takes 5 minutes to get things deliver to your home. No need to plan what snacks to order in advance. That’s the AHA.
Another friend completely stopped grocery planning. No more budgeting, no more meal prep anxiety. His cook comes in the morning, they discuss breakfast and lunch, and groceries appear in 5 minutes. Imagine the reduction in cognitive load. No planning the night before, no stress. Just wake up and decide what you want to eat. The groceries arrive while your cook is kneading the atta. It’s amazing.
Then there’s Superhuman. Their AHA moment was different - it was about realizing you could use shortcuts for the fastest email experience ever. But they didn’t wait for users to discover it. They made you sit through a mandatory 1-hour onboarding. Sometimes you can force the AHA moment. It’s more extrinsic than intrinsic, but it works.
I remember our obsession with finding our North Star when building a Slack-like enterprise app in my previous product role. We thought if people sent X messages, created Y channels, and added Z colleagues on day one, they’d reach their AHA moment faster. We kept adding hooks to drive these actions.
But here’s what we missed: correlation or causation? People who love your product will do these things anyway. You don’t need 50 hooks in the app. Plus, Slack already existed. Was there really a strong reason to choose us except pricing?
When Slack first launched, they had a genuine AHA moment. Enterprise communication tools like Yammer were pretty bad. Everyone was stuck with emails and endless sync meetings. Slack changed that.
But the 5th Slack clone? Probably not.
Same with Nubank in Brazil. Banking there had terrible NPS scores. They came in with an amazing experience - clear AHA moment. But if you’re launching a new online bank in India today, what’s your AHA moment? Why would people go “this is so much better than what I had before”?
Sometimes what feels like a positioning problem might actually be an existential one. Not every product needs to change the world. But if you’re chasing that AHA moment, make sure it’s real.
2024-12-18 08:00:00
The future of content and AI is going to be wild, and nobody is talking about it enough.
I was watching this episode of the My First Million podcast, and they were talking about these 1-minute serialized video reels. You know, the kind you see all over Instagram. Basically, TikTok dramas. I’ve seen a ton of them, and I know they’re huge in China. I know people who’ve actually paid to watch them.They’re always about some crazy, over-the-top storyline. Like, the boss falling for his secretary, or someone getting dissed by their rich relatives. But then there’s a twist – the guy getting humiliated turns out to be a secret billionaire. Forbidden love, revenge, the whole nine yards. It’s basically like those Chetan Bhagat novels or those Ekta Kapoor TV shows, but instead of stretching a scene for half an hour with five different camera angles, the story moves at lightning speed. Same actors, different shows. It’s wild.And it got me thinking. With AI video generators like Sora and Kling getting better every day, it’s not hard to imagine some dude in Hanumangarh pumping out these videos and making bank. I mean, yeah, you need consistency with the characters and more control over the video instead of just generating a single shot, but this feels like the first big way to make money with video generation. Forget about making entire movies – that’s probably still a decade away. This is something that could happen much sooner.
Speaking of regional content going global - Black Myth: Wukong showed that if you take a regional story and execute it well, the whole world will pay attention. I won’t be surprised if we see a AAA game based on the Mahabharata or Ramayana in the next 10 years. Sure, we might not have the game dev talent today, but as AI makes game development cheaper and faster, these IPs become more accessible. Just imagine playing as Arjuna in the Kurukshetra war or Hanuman leaping across the ocean to Lanka.
The education space is also going to change dramatically. Forget about White Hat Jr with their call center teachers reading from scripts. Imagine AI tuition centers where both your teacher and classmates are AI. The AI teacher adapts to your learning style, your AI classmates ask questions you’re too scared to ask, and nobody judges you for being slow.
You will have your copilot for interviewing who will be managing your own personal job-hunting army. They will help you through the entire application process. You tell your copilot what kind of jobs you want, and it orchestrates everything - one agent crawls job sites, another tailors your CV, another one applies, and another handles interview scheduling. The whole “spray and pray” approach to job applications will be replaced by targeted, personalized outreach at scale.
VR is about to get interesting too. The Oculus 3 costs as much as a cheap Android phone now. Meta is desperate for content. There’s never been a better time to build VR apps. Imagine a game where you’re Oberyn Martell fighting the Mountain in the Colosseum. (Though hopefully with a better ending for Oberyn this time.)
The future is going to be weird. But it’s also going to be full of opportunities. And I, for one, am excited to see what happens next.
2024-12-15 08:00:00
I have been using advanced voice AI a lot lately. Like everyone else, I am fascinated by how natural the conversations feel. But there is this weird UX problem that keeps bothering me.
You are having this deep conversation with the AI about some problem you are trying to solve. The AI gives you a bunch of steps or recommendations. Now you want those steps in text form because who wants to listen to a 5-minute-long response again and again?
Your only option? Exit the voice chat, go to regular chat, and ask the same question again. But wait, you were in the middle of brainstorming something. Now you have to start a new conversation. Explain the whole context again.
It’s like talking to someone who has the memory of a goldfish. “Hey, remember that thing we were discussing?” “Nope, who are you?”
I mean the Advanced Voice is really cool for chatting, but it kind of misses the mark when you actually need it to be, you know, useful. It’s more like a conversation buddy than an assistant.
I feel the ideal product would let you switch between voice and text seamlessly. Having a conversation but need something written down? Just ask for it in text. Want to continue the discussion? Pick up right where you left off.
The real dream is having an AI that can actually execute those steps for you. But for now, I would be happy with an assistant that does not force me to choose between voice and text. Add web search to this and we might actually have something interesting. Which I am guessing is coming soon anyway.
The more I think about it, this could even be a separate product. An AI that lets you have natural conversations but also sends you text when needed. Like that friend who follows up your deep discussion with bullet points on WhatsApp.
P.S. Every time I write about AI products, I wonder if six months from now this post will look silly because someone has already built this. But hey, that’s the fun part about writing about tech, right? Everything becomes outdated so fast.