You’ve probably noticed that every tech company is now an ‘AI company.’ They’re showing off chatbots that write poetry, image generators that paint like Van Gogh, and voice assistants that crack jokes. It’s flashy, it’s fun, and it’s mostly a distraction.
Because while the world is obsessing over which AI can pass the Turing test, a16z just quietly invested in a different kind of revolution. One that’s happening in the back offices of tax firms, on construction sites, in nursing homes, and inside agricultural supply chains. These are the places where the real money is—and where the real AI moats are being built.
Let me show you what I mean.
The Dirty Secret of AI Value
Look at the startups in a16z’s latest Speedrun batch. They’re not sexy. Grove Tax builds AI for tax preparers—the people who chase clients for W-2s, re-enter data into spreadsheets, and prepare filings. Piper-ai helps construction contractors read thousands of pages of bid documents to find hidden risks. Concorda is an ‘operating system for trial lawyers’ that automates the entire litigation workflow. Quanto does accounting grunt work.
The most valuable AI isn’t the one that writes your novel—it’s the one that does your data entry.
Think about that. The most advanced technology of our time is being aimed at the most mundane, repetitive, soul-crushing tasks. And that’s exactly why it works. These industries have been ignored by traditional software because they’re too complex, too fragmented, too dependent on human judgment. But AI doesn’t need to understand everything—it just needs to automate the 65% of time that tax professionals spend on non-judgment work. That’s the number Grove Tax cites: 65% of a tax preparer’s day is wasted on ‘clutter.’
From ‘Answering Questions’ to ‘Doing the Work’
This shift is fundamental. Early AI was about answering questions—chatbots, search, content generation. The new wave is about executing tasks. And that changes everything about what’s needed to build a defensible business.
Take Bilrost, which handles commercial lending. It doesn’t just read loan documents; it creates a ‘context graph’ of every transaction, connecting tax forms, financial data, borrower history, underwriting notes. It’s not a glorified search tool—it’s a system that processes the entire loan lifecycle from application to monitoring.
Or look at Vereda in Brazil, which uses WhatsApp to aggregate small farmers’ purchasing power and negotiate with suppliers. It’s not a fancy app—it’s a bot that does the grunt work of price comparison and order consolidation.
The companies that win in AI won’t be the ones with the best model. They’ll be the ones with the best data on the messiest problems.
This is why infrastructure matters more than ever. a16z also invested in companies that don’t do the primary work but enable it: Sentra builds ‘organizational memory’ for enterprises by capturing every meeting, email, and agent interaction. SafeWorld tests robots for safety in unpredictable human environments. Modaic validates AI decisions by assigning confidence scores. And Alike helps multiple AI agents coordinate, share information, and maintain privacy.
These are the plumbing, the wiring, the boring stuff. And it’s absolutely essential. As AI agents become more autonomous, the demand for memory, permissions, verification, and safety will explode. The small bets on infrastructure today will be the big winners tomorrow.
AI Steps Out of the Screen
Perhaps the most daring ventures are those that leave the digital world entirely. Smart Bricks is building an AI system for real estate investment that analyzes market data, evaluates properties, and manages assets. Syncere is making a home robot that looks like a lamp—until it sprouts arms to fold laundry. Clair Health is a wearable that continuously monitors hormone levels for women. Quo Labs is an AI caregiver for the elderly, starting with a simple voice assistant that reminds them to take medication.
These are hard. Hardware is hard. Regulation is hard. Liability is hard. But the barriers to entry are also the barriers to competition. The moment an AI system touches the physical world, its moat becomes as real as the concrete it’s processing.
And yes, consumer AI isn’t dead. Oasiz is building a ‘TikTok for AI-native software’—a social platform to discover and play with AI apps. PicPet is a virtual pet messaging app with 240,000 daily active users. Snag is an AI-powered sublet marketplace for Gen Z. But these require exceptional distribution, retention, and network effects. The bar is high.
The Real Takeaway
Here’s the twist: the AI revolution isn’t about making humans obsolete. It’s about making humans more human. By automating the robotic parts of our jobs—the data entry, the document review, the repetitive coordination—we free ourselves to do the work that actually requires judgment, creativity, empathy.
So the next time you see a demo of an AI that can write a sonnet, ask yourself: how much is that really worth? Then look at the startup that’s automating the billing process for a nursing home. That’s where the future is being built—one boring, indispensable task at a time.
AI is not coming for your job. It’s coming for the parts of your job you hate.
FAQ
Q: Aren't consumer AI companies like ChatGPT more valuable than these boring B2B startups?
A: ChatGPT is a phenomenon, but it's also a commodity. The real moats are built in verticals where data is messy, workflows are complex, and customers are willing to pay for outcomes. Boring B2B AI startups often have higher switching costs, recurring revenue, and less competition.
Q: What's the practical implication for a founder or investor?
A: Stop chasing the hype around general-purpose AI. Instead, find a messy, unsexy industry—tax, construction, insurance, agriculture—where human labor is bottlenecked by repetitive tasks. Build a system that automates the 'dirty work' and you'll have a defensible business with deep customer relationships.
Q: Isn't this just another form of software automation? What's different about AI?
A: Traditional software requires structured data and rigid rules. AI can handle ambiguity, natural language, and unstructured inputs—like reading a contract, understanding a whispered instruction, or interpreting a doctor's note. That's what makes it capable of targeting the 'hard stuff' that previous automation couldn't touch.