Yuta Baba

@levie this is so spot on. and this is why the visibility into the cost is more important than ever. you might want to even look into the customer level to see which customers bring good margin and bad margin

Nailing AI Agent pricing is a super important topic for AI companies right now. There are two dynamics at play in AI pricing land. The first dynamic is that the models themselves are getting cheaper and cheaper to run. But the other trend is that the use-cases from customers are requiring more and more inference. We’ve seen examples of Deep Research using up to 100X the compute of a standard query before. AI coding agents similarly can consume enormous resources depending on the task. So even as the inference gets cheaper per token, the total inference goes up dramatically. A normal thing to do with resource pricing is to effectively shift the cost of the resource onto the customer. It’s a clean model, but there are many areas where a key customer use-case may be technically possible but unaffordable today — even though they will become affordable tomorrow. So do you wait to solve the problem when it’s economically practical to scale, or lean in now and bet on the costs improving? The answer probably would be different in any other technology category in history. But the implications of AI model efficiency improvements is that software companies can afford to price AI in a way that anticipates the cost curve over time. This allows you to unlock use-cases today that may be otherwise less economically attractive but where you know they soon will be. It’s definitely a bet, but one that increasingly seems like it will pay off. This is all thanks to the constant AI breakthroughs coming from the frontier labs as well as open weights model providers. And this doesn’t appear to be slowing down anytime soon.

@AITECHio 100% agreed on this. That's why you have to measure the cost / ROI excessively. Love to chat if anyone is interested

40% Of AI Spend Goes To Compute! Most AI companies are not constrained by ideas. They are constrained by cost. According to McKinsey research, 25–40% of total AI spend is allocated to compute infrastructure. Not growth. Not hiring. Not distribution. Infrastructure. And it compounds. Every model call. Every user request. Every workflow. The more your product works, the more expensive it becomes to run. This is where most teams lose control. AI does not fail because the model is bad. It fails because the cost to operate it at scale breaks the business. The companies that win are not just building better AI. They are controlling the cost of running it.

@LearnOpenCV Super interesting! AI bill is going to get huge, totally agreed. We decided to build an observability layer for AI cost. Would love to chat if you are interested!

Your AI tool costs went from $20/mo to potentially $500K/quarter. And most companies haven't updated their budgets yet. Here's why AI agent billing is the next enterprise crisis 🧵👇 1/ A year ago: AI = autocomplete. Quick prompts, quick answers, flat subscription. Budgetable. Today: engineers hand AI entire codebases. 50-file PR reviews. Full microservice refactors. The meter never stops running. 2/ The unit economics are brutal: One AI code review ≈ $25 in compute. The agent runs a recursive loop — read → generate → execute → repeat. Every cycle costs money. The longer it thinks, the more it burns. 3/ Now scale it. Uber: 95% of engineers use AI monthly. Heavy agentic workflows. Not autocomplete — continuous labor. At enterprise scale, AI spend behaves like payroll, not a software license. 4/ The counterintuitive part? High AI bills = product-market fit. Nobody racks up $500K API bills on useless tools. A $50 compute burn replacing 2 days of senior engineer time is still a massive win. 5/ The next competitive advantage isn't the smartest model. It's FinOps for AI agents: → Route cheap tasks to cheap models → Reserve frontier models for complex reasoning → Real-time spend monitoring + kill switches The winners will be the best financial operators, not the best prompt engineers. #AI #AIAgents #FinOps #EnterpriseTech #AICosts #DevTools #Engineering

🚀 @carrotlabs__ai launched! Custom Models Tuned to Your Tasks That Never Stop Improving "Continuous Learning for Production AI." 🌐 fondo.ai/412Qelz Congrats @ChrisAcker10 & @yuta__baba!!

@mirandanover congrats!!

Introducing Fort, a wearable that automatically tracks strength training. Strength training is one of the best things you can do for your health and longevity. It deserves better tools.

@collinpounds @ycombinator @carrotlabs__ai Yes hence your proprietary models for your use cases that you own

@ycombinator @carrotlabs__ai Truly believe custom, fine tuned, local running models on M4 and M5 chips will win the race. Nobody wants to depend on internet and big tech for their super-intelligence to answer questions and no one likes sending their data off to be trained on.

Carrot Labs (@carrotlabs__ai) builds custom models tuned to your tasks that never stop improving. Their platform captures your agent activity, tunes a custom model, and continuously retrains as more flows in. The smartest model for your business is the one trained on it. Congrats on the launch, @ChrisAcker10 and @yuta__baba! ycombinator.com/launches/PdG-c…

@_Suresh2 @ycombinator @carrotlabs__ai 🥕🥕🥕

@ycombinator @carrotlabs__ai Custom AI models trained on your specific workflows is a game changer

@khaledealy @RishabhP821 @ycombinator @carrotlabs__ai let's say you want to process thousands of resumes and extract & classify info correctly. instead of balancing the consistent quality of opus, and the speed and cost advantage of haiku, you can have the best of both worlds with a model fine tuned on this task

@yuta__baba @RishabhP821 @ycombinator @carrotlabs__ai Can you give examples of specific tasks and who are the potential users?

@LeeLeepenkman @ycombinator @carrotlabs__ai thanks!

@ycombinator @carrotlabs__ai Nice congrats on launch

@Pericarus @ycombinator @carrotlabs__ai thanks!

@ycombinator @carrotlabs__ai This is neat

@beenwilli @ycombinator @carrotlabs__ai thanks!

@ycombinator @carrotlabs__ai Nice job! Congratulations

@readysetgopal The video!!!

Love a lot of things about this launch. First, the video is fire. Second, custom task specific models feel like they can unlock smarter agents for us

@RishabhP821 @ycombinator @carrotlabs__ai For specific tasks, you’ll be surprised

@ycombinator @carrotlabs__ai custom models, huh? what kind of cost are we talking to get a legit edge over gpt4?

@Kyriakos_Pelek @ycombinator @carrotlabs__ai Speed is the moat

@ycombinator @carrotlabs__ai YC startups moving fast again

@charu1603 @ycombinator @carrotlabs__ai @ChrisAcker10 Thank you!

@ycombinator @carrotlabs__ai Way to go @ChrisAcker10 and @yuta__baba! Great video! 💜

@AIWorkflowGuide @ycombinator @carrotlabs__ai create your own moat!

@ycombinator @carrotlabs__ai The data flywheel is the only real moat left.

@jrmyjlee @ycombinator @carrotlabs__ai thanks!

@ycombinator @carrotlabs__ai congrats guys!

Finally launched on YC!!

@ycombinator @carrotlabs__ai Let's goo!!