Why I want Yutori Running my Household and Business.

It’s 9:50am on Wednesday, March 19. My phone alarm goes off. Not because I had the chance to sleep in, but because I needed to remind myself to get ready to log into my children’s school portal at exactly 10am, click through 3 intermediate screens and book a time for parent-teacher conferences, all so that I could get a slot that would fit my schedule. After all, about 40 other parents were competing for the premium morning slots that occurred right after school drop off.

Sounds easy? Well, I have 2 kids. So I also had to get my husband to log onto the same portal simultaneously and book something non-overlapping for 1 kid while I did the other. We sat side by side so that we could verify this.

I run my own VC firm. My husband is an AI startup founder. The most stressful part of our day that day was from 9:50am to 10:02am. Seriously - the adrenaline was pumping.

Sound familiar?

It was somewhat gratifying, though, that over breakfast today — literally the day of said parent-teacher conferences — I found hope that my life might change soon. Yutori, a company I backed over nine months ago in their Pre-Seed round, was finally coming out of stealth!

Yutori is reimagining how we interact with the web. We all spend a significant portions of our lives online, doing soul-sucking, repetitive tasks. Whether it's booking conferences, ordering groceries, or coordinating a group trip — everyone needs (and deserves) a personal chief-of-staff. Yutori’s mission is to build the best AI assistants to tackle these problems and make space for what really matters — creating yutori (Japanese for “room to breathe”) in your life.

Yutori (ゆとり) is a Japanese concept meaning "room" or "leeway"—it refers to having mental, emotional, or temporal space to breathe, reflect, and live without constant pressure or urgency.

In a world of AI hype where many companies raise capital based on X demos, I saw my first glimpse of Yutori’s product just 1 month ago - 8 months after investing! But that’s because Devi Parikh, Abhishek Das and Dhruv Batra are actually trying to solve the problem. And it’s hard.

The fact of the matter is that many AI agents really don’t work all that well today, especially on the open Internet. Today’s frontier AI models enable great conversational chatbots but can’t complete tasks autonomously. That’s because AI Models still have a hard time doing simple things like navigating websites, recognizing what are pictures vs. buttons, understanding what information has to be filled out, reasoning through an entire workflow and doing multiple processing steps simultaneously. As the Yutori team once said to me: “Web agents are like a fleet of autonomous cars navigating the web, and require a similarly specialized stack.” The web is messy, non-deterministic, dynamic, and noisy. Mistakes are inevitable. Many current agentic applications actually compound errors rather than correct or prevent them. A cool demo for a specific scenario is one thing - getting it to actually work, across many use cases, every time, is something else, entirely.

So, what’s the solution? Scale compute?

While scaling seems to be the “market” solution for getting underlying LLMs to work better, this alone will not solve this problem. Most of what we have seen to date with LLMs can be classified as instinctual, gut reaction responses to queries we make - traditionally called “System 1 thinking”. It’s like the medical student training for the Boards - trained to spit out answers based on what it has learned / memorized. Impressive - and somewhat useful. But doesn’t really improve our health.

We want more. We want AI to help us think and solve problems and ideally, do real work on our behalf. We want the doctor who synthesizes information, creates a care plan and provides a recommendation. For this to happen, AI models must be capable of reasoning, problem solving and making informed decisions, or what we often call System 2 thinking. This has long been recognized by the research community but has been difficult to achieve.

Until now.

Yutori’s core innovation lies in bringing System 2 thinking (search, planning, and action) to web agents to augment System 1 (current LLMs) learnings. System 2 figures out what is the optimal action trajectory before executing an action - not after, as is the case with LLMs. So System 2 improves upon System 1 learnings … and then supervises System 1.

I could talk about tech all day. But as I always say: I back founders, not just ideas. Devi, Das, and Dhruv are a special team. They want to solve big, everyday problems for real people, all over the globe — not just the top 1%.

At Axiom, we call this “AI for the Real World.” That means building for 8 billion people, not just the 80 million living on the edge of innovation. Sometimes, as VCs, we are guilty of funding the things that only affect a small percentage of us.

Ok, the truth is, I actually do want Yutori for myself. I want it running my household — and large parts of my business too. Because it’s dead simple to use, and it actually works. A value proposition that resonates with, well… everyone.