Really interesting breakfast this morning with the @amazon and @AWS robotics teams and 15 early to mid stage venture investors. Lots of excitement around the potential for automation, mixed in with serious skepticism around ROI for many use cases going one level deeper. What's the cost of the human or humans + machines doing the thing now and what is the complexity (and therefore cost) of the robotic application that will replace them. Stationary arms are deployed in manufacturing at scale, so we know the ROI works there.

Biggest takeaway from @ycombinator yesterday was the number of founders building in hardware and physical AI. @garrytan mentioned that 25% of the batch was hardtech. Part of me wondered whether the spotlight on physical industries was eroding an “earned secret” we’ve had at @nvpcap after investing in the space for the past two funds. Maybe the secret is out. But after hearing the pitches, I walked away with the opposite reaction. The quality of the founders was incredibly high, 10% of the hardware founders had PhDs, and the mission orientation was front and center. It felt less like the opportunity is getting crowded and more like the pie is expanding. And there are still plenty of investors who won’t lean into the physical world because it’s hard and sits outside their traditional pattern recognition.

This is a great take from @shomikghosh21. linkedin.com/pulse/building-…. What really stood out for me was 'Deloitte projects US manufacturing could need as many as 3.8 million workers between 2024 and 2033, with up to 1.9 million of those roles going unfilled. 65% of manufacturers told the NAM that attracting and retaining talent is their single biggest business challenge.' That is HC that either has to be trained or filled with automation.

This is a great earned view of the balance between robotic deployment and training data. And on the training data front I agree that 'The important question then isn’t about the usefulness of data, it’s about opportunity cost and uniqueness. What data provides the highest ROI and where is there greatest or least flexibility in the training mixture?' waldenrobotics.com/news/auto…

In recent conversations on AI robotics with our founders and researchers in the space, two ideas stuck with me. 1) AI enabled robotics seems to be following a different path than software only: instead of building on top of a small group of dominant model stacks there is a broader set of platforms and meaningful value being created at the application layer for companies that lean into deploying into production early. 2) The most compelling near-term companies may be the ones that pick a specific vertical, solve a real problem with today’s capabilities, and then scale from an actual product rather than a lab demo. That’s why I’m increasingly excited about robotic applications in areas like inspection, logistics, manufacturing, and other mission-critical workflows where automation can deliver value before we get to “general-purpose everything.” The infrastructure and model layers will keep evolving, but robotic application companies that are tightly aligned to customer pain and operational reality may be in the best position to build enduring businesses. The integration into physical environments, customer workflows and operations creates switching costs that pure software doesn't have.

💡 One final thought. Your first AE is a foundational team member, similar to a founding engineer, so I'd spend more time recruiting them, pay them more than feels comfortable, and treat them as one of the most important early hires in the company.

6️⃣ Hire AEs before a Head of Sales. I've seen it done both ways, but hiring a head of sales first commits you to a scaling path without all the information you need. There are exceptions, especially if demand is overwhelming, but most founders benefit from starting small.

5️⃣ If you can afford it, hire two. Two AEs give you an experiment. If one succeeds and one struggles, you've learned something. If both struggle, maybe the issue isn't the people.

4️⃣ Don't overvalue network. Everyone loves the candidate who "knows everyone." A network is helpful (and in some industries like defense, it is a non-negotiable), but warm introductions run out, I'd rather have someone who knows how to translate a company vision into sales.

3️⃣ Pay them like a builder. Your first AE isn't just selling, they're translating the your vision into language customers understand. It's a core team member like an early engineer or product person and so I'd pay them accordingly, in cash and equity.

2️⃣ Hire someone who wants to build the playbook. Your first AE isn't your fifteenth. The best early AEs tend to have operated in ambiguity before. They've built something from scratch instead of simply executing a well-oiled machine.

1️⃣ Don't hire your first AE until you've proven some level of real customer value. Founder led sales should prove demand, and ideally a handful of paying customers as ideal profiles for the future pipeline.

We've watched 40+ B2B startups hire their first AE. Some hires worked. Some didn't. Here are the 7 patterns we've seen. 🧵

My most popular essays of the past year are: - The Electric Slide: 40k words on the history, geopolitics, strategy, and technical details of each layer of the electric stack. - a16z: The Power Brokers: a 16k word essay on a venture capital Firm on the day of their $15B raise - Riding the Leopard: a talk on the meaning of life citing sources from Jesus to a guy who took a ton of LSD. Not sure what the through line is other than these are probably the three essays I poured the most effort into. I feel very lucky that I can write about whatever I find most interesting at any given time and if I’m curious enough about it, people seem to like it. Essays below.

Sharing a great piece from Musings from the Factory Floor blog. Lots of new manufacturing (commercial and defense) funding in the past couple of years, which is a precondition for U.S. industrial leadership - but there is still a lot more to do. americanfactory.substack.com…

Being a @nyknicks fan in the 2000s was non-consensus…and finally right.

The @cerebras IPO is a win for their employees and investors. I'm neither! But I learned 2 things as a seed investor: 1) Cerebras was non-consensus in 2015 when it started, 2) The initial offering 2 years ago at 1/10 the price didn't go. Time + tailwinds can change a lot.

We're seeing a ramp in early stage acquisitions, which was once a staple of venture-backed exits. Some of the same players are involved (Google, Meta), but now OpenAI, Anthropic, SpaceX and others are in the mix too. That's a positive for the ecosystem. techcrunch.com/2026/05/01/me…

I wrote about robotic automation in January, and a concern was historically slow development in the space. But, acceleration in funding in world models + training data for physical AI in the last few months is impressive - time to update the thesis! nvpcap.com/blog/physical-ai-…

Focus on the 20% and be positive!