> ## Content Index
> Fetch the complete content index at: https://longtalkshort.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Robots by the Hour: Tutor Intelligence's Pitch to Factories That Have Never Bought a Robot
- URL: https://longtalkshort.com/tutor-intelligence-josh-gruenstein-robots-by-the-hour-a16z/
- Published: 2026-11-03T17:30:25.000Z
- Updated: 2026-11-03T17:30:25.000Z
- Description: Tutor Intelligence CEO Josh Gruenstein says most US factories have never bought a robot. His pitch: general-purpose robots rented at about $16 an hour, with remote humans whose help trains the AI. a16z, which hosted, led the company's Series B.
- Author: Truman
- Tags: Economics, Josh Gruenstein, Martin Casado, a16z

**You'd assume a robot boxed the last thing you bought. Probably not.** Josh Gruenstein, co-founder and CEO of Tutor Intelligence, says nearly everything that arrives in a box was packed by a person, even at iPhone scale. In an a16z episode with general partners Martin Casado and Raghu Raghuram, he argues that the big robotics opportunity isn't a humanoid in your kitchen. It's the ordinary factories and warehouses that have never bought a robot, reached by renting robots by the hour with humans ready to help them. **One thing to know up front: a16z led Tutor's $75 million Series B, so the hosts are investors in the company they're interviewing.**

*In short: Gruenstein says most US factories and warehouses have no robots because traditional automation is custom, slow and expensive. Tutor sells general-purpose robots as labor, pitched at about $16 an hour, with remote human "tutors" who step in when the AI gets stuck. That human help becomes training data. The pitch is plausible, and US robot installations do look small, but many of the figures are the company's own.*

## 1\. The market is mostly empty

Gruenstein's opening statistic is that US companies installed about 38,000 industrial robots in 2025\. That sounds like a lot until you compare it with the number of workplaces. **In his telling, nine out of ten US factories and warehouses have no robots at all.** "Almost everybody we're selling a robot to has never purchased a robot before," he says. "Almost all of the market is greenfield."

Why so empty? Traditional industrial automation is built like a custom machine: engineers design a cell for one task, at one site, over months, and it pays off only at high volume. **A mid-sized co-packer that changes products every few weeks can't justify that, so the work stays manual.** He describes jobs that are backbreaking and hard to hire for, with three-month turnover and an average worker age of 55\. So how do you sell to a buyer who has never bought a robot?

## 2\. Sell labor, not machines

Tutor's answer is to stop selling robots as capital equipment. Its sales team doesn't lead with AI:

> We call them up and we're saying, 'Hey, have you ever considered a robot for 16 bucks an hour?'

**That framing puts the robot next to a temp worker in the buyer's head, not next to a capital budget.** Gruenstein says first call to signed contract averages 40 to 60 days, with robots working on site about 40 days after that.

Making it work at that price means owning the whole stack. Tutor designs and manufactures its own robots: Cassie, a single arm with suction cups for heavy picking and packing, and Sunny, a wheeled semi-humanoid. Off-the-shelf industrial arms, he argues, were built for fast repetitive motion, not for running AI models. The company has about 160 people, mostly in the Boston area.

![A side-by-side comparison. Left, a traditional custom automation cell: one task, months of engineering, a large upfront cost, built for high volume. Right, robots rented by the hour: general-purpose robots, quick deployment, an hourly price, suited to small and mid-sized sites. No people are shown.](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/11/fig1-16.png)

Gruenstein's contrast between custom automation and robots sold as hourly labor. Our grouping of his argument. Made for Long Talk Short.

## 3\. The human in the loop is the data engine

The company's name explains its approach. Language models learned from the internet, a vast record of human writing. There's no equivalent for physical skill: no one recorded the pressure of every fingertip when someone picked up a coffee cup. **So Tutor creates that data. When a robot on a customer site is unsure, a remote human tutor guides it, and those corrections feed back into the models.**

The argument in the episode is that this data beats lab data because a customer is paying for the work. Data from a lab tells you what's possible; data from paid work tells you what's economically valuable. Tutor also runs "Data Factory 1", a fleet of 100 Sunny robots in Boston collecting data for research, which Gruenstein calls the largest semi-humanoid data farm in the US to the company's knowledge.

![A loop diagram: robots at customer sites do paid work; when stuck, a remote human tutor helps; the corrections become training data; better models let robots handle more tasks; more tasks bring more sites. No people are shown.](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/11/fig2-16.png)

Tutor's described flywheel. Our drawing of the company's pitch, not a measured result. Made for Long Talk Short.

## 4\. Why build everything in-house

Casado presses on whether owning hardware, models and deployment is a phase or permanent. Gruenstein's answer echoes early carmakers: **when a market is immature, nobody sells the parts you need, so you build them yourself.** Once volume is big enough, the industry can re-specialize, and suppliers become something closer to contract manufacturers. Until then, owning the stack lets Tutor change its manufacturing process every two weeks and ship software to robots daily.

The deeper bet is that the advantage comes from the flow of data, not clever architecture. **Any technical trick is worth doing, he says, only if it helps deploy more robots, collect more data and take on more tasks.**

## Weighing it: what holds up

**Framing:** The iPhone packaging line in the episode title is a single remark, not the episode's theme. The conversation is a company pitch about selling robots as labor, and we've framed it that way.

**Holds up:** **The International Federation of Robotics reports US industrial robot installations of about 38,400 in 2025, up 12%, which matches Gruenstein's 38,000.**

**Partly holds:** IFR counts about 354,000 robots installed in China in 2025, more than half the world's total. That cuts against his point that China will "have to tear down" its factories while the US builds fresh: China is already automating at a far larger scale.

**Corrections:** His claim that Tutor robots have helped ship "over a quarter of a billion products" is an estimate he says came from asking Claude, an AI chatbot; it isn't an audited figure.

**Not checked:** that nine in ten US factories and warehouses have no robots, the average worker age of 55, three-month turnover, and the 40-to-60-day sales cycle; treat them as the company's figures. Cost-savings examples in the episode are illustrative, and the robots' success rate without human help wasn't disclosed.

**Disclosures:** **Andreessen Horowitz led Tutor Intelligence's $75 million Series B, bringing its total funding to about $117 million.** Casado mentions talking to Tutor's customers "before we invested". The a16z hosts are investors promoting a portfolio company, and the episode should be heard as a pitch. We worked from YouTube's automatic captions, so quotes may contain small errors.

**Bottom line:** the empty-market thesis and the hourly pricing are credible ideas, and the US installation figure checks out. The growth and deployment numbers come from the company, with its lead investor asking the questions.

**Related reading:** [a16z's American Dynamism list and why production is deterrence](https://longtalkshort.com/a16z-american-dynamism-50-production-is-deterrence/) · [Ethan Mollick's four rules for working with AI](https://longtalkshort.com/ethan-mollick-co-intelligence-worst-ai-yet-four-rules/)

## Long Talk Short

- Most of what arrives in a box was packed by a person, because custom automation rarely pays off for small and mid-sized sites.
- US robot installations are small relative to the number of workplaces, and far below China's.
- Pricing robots by the hour reframes them as labor, not capital equipment.
- Remote humans who rescue stuck robots can generate the training data physical AI lacks.
- The growth figures come from the company, and its lead investor hosted the interview.

**The one thing to remember:** in robotics, the business model may matter as much as the robot.

*This summary is for general information only and is not investment advice. Do your own research or consult a qualified adviser before making investment decisions.*

Source: a16z with Josh Gruenstein, Martin Casado and Raghu Raghuram, "What iPhone Packaging Reveals About the Robotics Opportunity," *a16z*, on YouTube. This is an independent summary and is not affiliated with a16z or Tutor Intelligence.