Yalo

Day 62 of building Sara, my personal AI agent. There’s an idea that’s been circulating in tech for months, and it came up again in Sequoia’s AI Ascent 2026 keynote ( https://lnkd.in/eB_AS4Zq ): AI is not just expanding software. AI is expanding software into services. Sequoia frames it clearly… In the first 15 years of the cloud transition, the software TAM went from ~$350B to ~$650B. Services is a much bigger pie. US legal services alone is ~$400B. One vertical, in one geography, almost the size of the entire software market. The total services pie could be $5T, $10T, or more. For the last 20 years, software helped humans do work through dashboards, CRMs, spreadsheets, search boxes, buttons, and forms… But the actual work mostly lived in services. Assistants, analysts, recruiters, lawyers, accountants, consultants, support reps, tutors, coaches. The shift is that AI lets software cross into services… and Sara is my personal version of this. She is not “ChatGPT with memory and connectors.” She does services for me. Actual work. She summarizes what I’m learning, prepares notes, tracks reminders, drafts posts, builds prototypes, researches things, manages files, helps me think, and turns messy inputs into structured outputs. None of these are apps. They are jobs. Small jobs, but real ones. Once an agent can reliably do enough small jobs, the product stops feeling like software and starts feeling like labor. This is why I think the opportunity is not “AI tools.” The opportunity is applied AI services that become software over time: Start with a job. Do it manually or agentically. Learn the workflow. Add memory, tools, approvals, and taste. Make it repeatable. Then productize. For executives and founders, the question is not “how do I make my current team more productive?” It’s: what expensive service can now become software? What recurring job can an agent do 10x cheaper, faster, or better? What workflow was impossible to automate before, and is suddenly possible? That feels like the frontier. Not better apps or prettier dashboards. Software that does the work. And as Sara gets better, the sky is the limit on how many more services she’ll be able to do.

Day 61 of building Sara, my personal AI agent. Today I wanted to buy something very specific for my son: the hardcover Mexico version of the World Cup Panini album. Simple request… but shopping is rarely just about the product. Very quickly you run into a bunch of small questions: Where is it sold? Is the exact version available? Can it ship to my hotel? Will it arrive before I leave the city? Is it faster to order online or walk into a nearby store? What’s the total cost with shipping? This is what shopping still feels like today… Searching… Comparing… Asking questions… So I asked Sara. She found the official Panini Mexico store, identified the right hardcover version, checked the price, found my hotel from my calendar, confirmed the address, calculated shipping, and got the item into the cart. She got 90% there… but then she got stuck at payment. I’m still figuring out how to give Sara a wallet. I’ll probably start with something like Stripe Link + spending limits + explicit approval before purchases. But the direction feels very clear to me… The future of shopping is not browsing 20 links. It is delegating the job to a personal agent, especially when you already know exactly what you want. Your agent should know your preferences, your calendar, where you’re staying, how long you’ll be there, your delivery instructions, your budget, and when to ask for permission. Then shopping becomes: You ask. Your agent does the work. It asks for approval. You tap yes. Shopping moves from search to delegation. The next shopping interface won’t just be a search bar or a homepage... It will be an agent trying to complete a job on behalf of a customer. Companies should be working to make their catalogs agent-ready.