Sep 20, 2017 · 24m · 20vc
20VC: Why Core AI Is Largely A Bunch of Nifty Tricks, Why Consumer Electronics Is The Hardest Thing To Do In Venture & Why Silicon Valley Will Dominate The Future of Robotics with Jeremy Conrad, Partner @ Lemnos Labs
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The 20 Minute VC, host Harry Stebbings interviews Jeremy Conrad, partner at Lemnos Labs, to discuss hardware venture capital investing, robotics trends, financing milestones, and the operational realities of hardware startups.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 40.5% of the talking time here. How this is scored →
Harry as informed peer 4.0 Guest teaching 4.1 Guest disagreement 2.9 Harry pushing back 3.1
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 6:07 AI as nifty tricks and mechanical turks
Jeremy forcefully rejects Harry's premise that robotics requires AI/ML, calling current AI applications 'a bunch of nifty tricks' and 'a little bit mechanical turk'.
Hardest push from Harry ▶ 6:07 Harry directly challenging AI necessityHarry poses a provocative counter-question asking whether advances in vision and hardware are useless without core AI and ML, directly testing Jeremy's thesis.
Biggest teaching moment ▶ 6:07 Deconstructing AI/ML hype in productionJeremy educates Harry on the reality of current AI production, explaining how companies like self-driving car developers still rely on massive human teams manually tagging data.
Harry holds his own ▶ 13:24 Citing Reid Hoffman on minimum viable productHarry demonstrates strong tech industry expertise by citing Reid Hoffman's famous quote on product embarrassment to test the limits of hardware iteration speed.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Show Transition and Arrival Stinger | 2 | 2 | 1 | 1 | Harry introduces Jeremy and asks about his transition into VC and the inception of Lemnos Labs. Jeremy explains how cheap CAD tools, rapid prototyping, and Arduinos lowered hardware startup costs without any friction or pushback. | |
| Mega Trends Driving Robotics vs. AI and ML | 5 | 6 | 6 | 6 | Harry explicitly challenges Jeremy by asking if robotics strictly requires AI and ML to be useful. Jeremy forcefully rejects the premise, dismissing current AI/ML as a 'bunch of nifty tricks' relying heavily on manual human data tagging. | |
| Why Silicon Valley Dominates Robotics and Hardware | 3 | 4 | 2 | 2 | Harry inquires about Silicon Valley's dominance in hardware and robotics. Jeremy explains the 'Midwest Curse' research and the psychological drive in SV, while Harry probes into the hidden downsides of Bay Area founder culture. | |
| Financing Challenges and Series A/B Metrics for Hardware | 5 | 4 | 2 | 3 | Harry brings up Chris Duvos' question regarding financing challenges for Series A/B hardware companies. Jeremy outlines how Lemnos helps founders de-risk milestones and highlights the tightening requirement for real revenue at Series B. | |
| The Hard Truths and Viable Models of Consumer Electronics | 6 | 5 | 3 | 4 | Harry quotes Reid Hoffman on shipping imperfect products to challenge hardware iteration constraints and discusses failed consumer hardware like Juicero and Pearl. Jeremy explains why broken hardware cannot be shipped and details viable business models with recurring revenue. | |
| Bearish Thesis on Space Startups vs. High-Frequency Data | 5 | 6 | 4 | 5 | Jeremy details his bearish thesis on space imagery startups, citing low usage frequency and regulatory spectrum hurdles. Harry pushes back by asking if space capital demands are functionally distinct from standard hardware risks. | |
| Lemnos' Strategy as a Specialized Early-Stage Fund | 2 | 2 | 2 | 1 | Harry jokes about VC management fees before launching into a quickfire round covering books, killer AI, and favorite investments. Jeremy offers rapid responses in a highly collaborative and lighthearted tone. |
Statements from this episode (19)
VCs refused to fund hardware startups in 2011 due to high costs
“And we started to talk to some other people and really found out that, you know, no one was interested in funding hardware startups. And so we asked around, we said, well, why? And they said, well, it takes too long and it's too expensive.”
Cheap computing and 3D printing drastically reduced hardware startup costs by 2011
“Today, any 500 dollar computer can run incredibly sophisticated CAD. And on top of that, the software is Free. And then you kind of add in rapid prototyping like three D printing, and you're kind of timing for your first unit, plus things like Arduino, make it…”
Advances in sensors and compute drastically reduced robotics startup costs
“All of those things conspire to make the cost of starting a robotics company a fraction of what it used to be.”
Core AI and machine learning in 2017 is mostly just nifty tricks
“So I really don't think it does, and I would also argue that what people consider a core AI ML today is a bunch of nifty tricks, and so you see lots of nifty demos, but in terms of actual production AI or ML actually out there, a lot of it's a little bit mecha…”
Shorter hardware stacks make robotics accessible yet defensible for startups
“And hardware, the stack for a hardware company is shorter than it's ever been. And so that's why I think it's a great time to start a robotics company, because it's hard enough that it's defensible, but easy enough for a startup to do.”
Robotics startups cannot bet on locomotion and motor costs dropping rapidly
“The two things that I don't think are really that you can't bet on that will get better are locomotion and manipulation. So if your robot needs to drive around, you can't assume that, oh, like three years from now, my motors will cost a lot less.”
LIDAR sensor prices will drop fivefold between 2017 and 2020
“Whereas if you have LIDARs in your robot today, You can assume that they're going to drop five X in price in the next three years.”
The best new robotics startups focus on a single obvious ROI
“I think the best robotics companies started today will have like a singular use case where the ROI is obvious and apparent.”
Social media companies unfairly dominate mindshare over Silicon Valley's hardware firms
“And there's dozens of incredible hardware companies here, but they just don't have kind of the mindshare that I think some of the social media companies do.”
Series A and B investors will back hardware if the business scales
“The reality is, is that they just want to invest in great businesses, and if you can convince them that this is a scalable, venture-backed business, they'll look at anything.”
Series B hardware investors now demand proven revenue rather than hype
“I think that two years ago, the ecosystem was in a place where you could go raise a big A or small B on like kind of hype and, you know, good team and like that, the idea that there was going to be a market there. But right now, you know, we're seeing a very m…”
Hardware startups must narrow scope because 90 percent functionality is completely useless
“I think it's about making sure that you really narrow your scope for your first product, because if 90% of your product works in its hardware, you have nothing.”
Consumer electronics is venture capital's hardest sector due to massive capital needs
“So I think consumer electronics is one of the hardest things to do in the venture ecosystem right now, partially because at some point you need a huge slug of money to go big.”
Selling standalone hardware without recurring revenue is no longer viable for startups
“So for us, when we look at consumer electronics, one, we think that single cell hardware is probably not viable right now. So there has to be some sort of recurring revenue model.”
VCs will stop funding consumer hardware companies before product-market fit
“But I think we're going to see a lot less of things like Juicero or Pearl, where they raise significant amounts of capital before finding out if they had product market fit.”
DJI is a $10 billion consumer electronics company showing no slowdown
“But also, let's be super clear, you know, DGI is just a consumer electronics company. It's a ten billion dollar company. And they're showing no signs of stopping.”
Commercial satellite imagery lacks high-frequency daily demand, making it a bad bet
“And so like, I'm pretty bearish on the kind of imaging side of it”
LEO satellite communications startups are a terrible fit for venture capital
“There are groups trying to do this with these LEO constellations, but the first thing they do is spend hundreds of millions, if not billions of dollars on spectrum, and that's not a good fit for the startup ecosystem.”
People misjudge exponential S-curves when worrying about killer AI
“I'm not at all worried about killer AI. I think that people love to talk about, you know, when you're on the early part of the exponential, you can't tell. And I would argue that when you're on the exponential part of the S curve, it's hard to tell as well.”