Software's Blind Spots
Hey there, investors and innovators!
The easy places to put AI have mostly been taken. This week's picks go somewhere harder: the courtroom, the repair bay, the employee's daily workflow, the film editor's timeline, the map on your phone. These are settings where the work happens in the physical world or in someone's head, and where almost nothing about it was ever written down. The common move across all five is to capture that work, turn it into data, and then build something that acts on it. Four of the five teams are two founders, and the fifth is a team of 80 that is already running at national scale, so the batch is showing both ends of the range this week.
What the Pattern Says
- The next wave of AI is capturing work nobody recorded. A repair bay, a court hearing, an employee learning a new tool: none of these left a clean record before. Several of this week's companies start by capturing the work itself and treat the resulting data as the real asset, with the product layered on top. That puts the moat in the data collected, not the model underneath.
- Concentrated buyers are producing fast early traction. Dealer groups and state court systems buy in large blocks, and the traction shows it: dozens of dealerships in a few months, statewide mandates in a handful of states. In markets where one decision reaches hundreds of sites, distribution is a smaller problem than it looks.
- Control is beating raw generation. In music, the pitch is hitting an exact beat on request, not producing a plausible track. In the workplace, it is guiding a person through a specific task rather than offering a generic chatbot. As generation gets cheaper, buyers are paying for precision and for results they can steer.
- Scaled operators are starting again with tiny teams. Founders from Cambly, Zenly, Life360, Navi, and Sarvam are back with two-person companies, aiming at problems their earlier teams needed far more people to approach. Experience building at scale is now being combined with a lean AI-native team from day one.
- A strong technical pipeline from India's engineering schools and AI labs shows up across the group. Founders trained at IIIT Hyderabad, IIT Kanpur, and BITS Pilani appear across three of the five companies, and one company is building for India's courts at national scale. The talent base building for the next wave of AI is increasingly global.
Eight Capital tracks every YC batch, cohort to cohort. Reach out if you'd like a deeper cut on any theme or company above.
Fund Manager Picks
Anywhere
Your places, plans, friends and memories, all on one living map
Why it's interesting
(San Francisco, 2 people): A personal, social map that pulls together the places you care about, your friends, your plans, and your memories, with AI agents to help you decide where to go. The founder has spent a career building maps and location products: he was an early employee at Zenly before the Snap acquisition, where he led the map effort, co-founded the Bump app, and was a principal engineer at Life360, where he rebuilt the map experience across iOS and Android with a two-person team. He also built a cross-platform C++ map engine from scratch.
Adalat AI
AI OS transforming courts, accelerating justice
Why it's interesting
(Bengaluru, 80 people): AI infrastructure for India's courts, with the goal of speeding up justice. The founding pair combines legal and technical depth: the CEO is a former Supreme Court of India law clerk with degrees from Harvard and Oxford, and the CTO led ML engineering at a prior startup and has published research on legal AI and Indic speech. The team says it is live in about 25% of India's courtrooms across 11 states, including statewide mandates in 3 states (company-reported).
ChampionEffect
An AI coach for every employee!
Why it's interesting
(San Francisco, 2 people): An AI coach for every employee, living inside Slack and Teams. It explains new AI concepts and tools, suggests the best use cases for a person's industry, and can join a call to walk someone through the work, then uses what it learns to help their coworkers. The founders previously built Cambly, a YC company that taught English to millions of professionals and grew past $100M in revenue, and one of them built its go-to-market organization from zero.
Taan.ai
the most precise & controllable foundation music models
Why it's interesting
(San Francisco, 2 founders): A music model built for precision and control. It reads a video's cuts, mood, and timing, then composes and mixes a score to match, and users can steer it with text, a hummed melody, exact notes, MIDI, or images, with downloadable stems. The company says it trains on licensed music. On its own 250-prompt timing benchmark it reports hitting timed events within half a second 73.2% of the time, against 6.8% to 12.8% for the named competitors (company-reported). The CEO is a touring musician and music-ML researcher with two patents on music generation architectures, and the CTO did signal-processing research at IIT Kanpur.
Bayworks
Intelligence that keeps America’s vehicles running
Why it's interesting
(San Francisco, 2 people): Voice and video capture of technicians as they repair vehicles, verifying the work and turning each repair (symptom, diagnosis, fix, outcome) into data that runs dealership operations and standardizes repairs to manufacturer specs. The founders bring voice AI experience from Sarvam and fintech engineering from Navi and Scapia. The company reports growing from zero to 80+ dealerships and about 4,000 repairs a day in 110 days, including a top-50 US dealer group with $3B+ in annual revenue (company-reported).
