Hardware rules.

I love hardware. Especially heatsinks. Not really sure why — they’re kind of boring, usually necessarily ugly (all those awful fins!) — but after designing and building one for a consumer product I tell you what, they’re tricky buggers!

I never forget a good, big heatsink.

And thats why, when I saw the news this morning that the US government has banned US companies from selling product or technology to a gaggle of Chinese AI companies — Hikvision, Sensetime, Megvii, more — visions of heatsinks surprisingly worked through the noise that is my stressed out, overcaffeinated brain.

Wait a minute…I have a Megvii camera in my garage! And it has an awesome heatsink!

I wonder whats inside?

Hey Andy — why do you have an Megvii camera used for facial recognition by the Chinese government?

Good question. It all began in around ’09, when a bunch of Portuguese…

But that’s neither here nor there — recessions are a bitch.

But see that little sticker next to the Megvii logo? The beautiful purple-hued brain? That’s Comet Labs baby, subsidiary of Legend Star out of Beijing and my former employer. Legend Star, turns out, was the lead Series A investor for Megvii.

‘Bout 8 months ago, Legend Star (rightly, because of CFIUS) decided to stop US investments through Comet. Its never fun to be given a cardboard box and told “get out and stay out,” and in a fit of entitled rage I grabbed the most expensive looking thing — a security camera — from a pile of unused stuff in the corner. Take that, Comet Labs. Take that.

So now you know — in some way, Megvii paid my salary and got me where I am today, working with Micron Technologies and their AI fund.

You know how the hot chicks in high school always fell for the bad boys? After writing that last paragraph, I kind of feel like that right now. Bad boys, why you make me feel so good?

Because I’m an extremophore, thats why. I live for risk, thrive in stressful situations, and love working with bad ass folk who know their stuff and aren’t afraid to go big. That’s startups. That’s Legend Star. That’s Micron.

Wolves hunt together.

Andy — who cares about you? What in the world you got in that camera!

Right. Taking this puppy apart, I learned three quick things:

1. its good Megvii have software valuations, cuz their hardware kind of sucks. Pieces were falling off left and right with only the smallest pull. Weird.
2. Thank God they didn’t use threadlock. I was expecting 2 hours — dismantling the thing took about 5 minutes.
3. Oh, that heatsink.

First thing I saw inside was the camera lens and a big sticker proclaiming “Made in Japan.” Not surprising — we offshored all our optics manufacturing capabilities decades ago. But I’m a Micron guy, and I’m not here lookin’ for some rinky-dink glassware. I want the good stuff. Semiconductors.

Heres the first board:

VIMICRO — never heard of ’em. https://en.wikipedia.org/wiki/Vimicro — “It was founded in 1999 when the Chinese government invited a group of Chinese people who had been educated and had established careers in Silicon Valley to return to China to start a company.” Well I’ll be.

SAMSUNG — heard of ’em, and that looks like memory to me. Thanks, Megvii. Thanks a lot.

But this is just image processing components — nothing too wild, and I haven’t yet cracked into that awesome heatsink. But again, cracking into that awesome heatsink was a function of four screws and kind of tapping the heatsink on the table — like an egg, but with screws. Easy.

Inside were two boards, positioned perpendicular to each other in kind of a “T”. The first, the top of the “T” and sunk to the rear of the heatsink, really just had a bunch of peripherals and isn’t worth the bytes to share. Boring.

But the last board was smashed in between the two halves of my great big heat sink egg.

The thing about heat sinks, they’re usually good at extracting heat. And again, this is a big heatsink, so whatever kind of stuff they got going on in these heat sink innards has gotta be working up a sweat.

Here’s the first side, the one without the annoying sticky stuff adhering it to the heat sink (while I love heatsinks, I love thermal interface materials more):

JORJIN — don’t know ’em (I need to know more people building this stuff). And they dont have a wikipedia page. They’re in Taiwan and build wireless modems. Got it. Yawn.

HYNIX — uhh, yeah thats more memory. Come on now, Megvii! Thats two out of the big three — whats a guy gotta do to sell some DRAM around here?

Still, nothing too crazy and nary a US component as far as the eye can see (well, ignoring that noise about VIMICRO, and maybe I missed one of the passives).

Then I flip the board over:

HYNIX — aaaarrrrrgggggg!!!!!

KINGSTON — Hmmm. That’s some American storage there.

And that gorgeous reflective processor in the middle?

NVIDIA — interesting.

With enough distance, any surface appears infinitely uniform.

The ocean? From our seats on the plane it’s just a bunch of blue. The Milky Way? Pfft — not even a grain of sand in the great big multiverse.

Thinking this way can be a bit terrifying — I stopped reading about the Big Bang after losing too much sleep staring at the ceiling, questioning the cosmic meaning of awkward BART rides — but it is an almost essential mindset for machine learning founders looking to build value.

Prediction and automation services working with surface level data result in undifferentiated offerings and step-wise value creation. More importantly, they miss out of the exciting things that come when you dive in, get dirty, and keep going deeper than ever before.

If you believe this — like we do at Comet Labs — it follows that startups creating the most value using machine learning technologies are those that enter markets with:

1. A minimum viable product based on some kind of automated service leveraging rich, industry specific data (i.e. fast entry and stickiness); and
2. A productization strategy that improves operational efficiencies over time (keeps getting better and better).

One of our recent investments — PingThings — is an excellent example of a strong, focused, warrior team not only solving industry-specific problems but creating original value by journeying deeper than ever before with early industry customers.

We invested in PingThings because they’re the best at what they do.

PingThings is valuable because what they do is challenging, tedious, and boring — and it sets their utility customers on a path to improve their operations by orders of magnitude and in previously impossible, unknown ways.

“We have so much data, we don’t know what to do with it”

Utilities. Auto OEMs. Appliance manufacturers. Governments. Insurance providers. Banks.

If I had a nickel for every time I’ve heard large entities say “we have the data, but we don’t know what to do with it,” over the past decade, I’d have about $3.35. Which isn’t a ton of money, but it is a lot of people reflecting on the real value data access has generated for their operations.

Take utility smart meters.

Once upon a time, when smart meters were first being deployed at scale, the conversation was about how capturing energy usage data at a finer scale would revolutionize utility operations. Fewer losses! Improved demand response! A higher penetration of distributed energy!

In reality, the dominant value of smart meters in industry is…wait for it…the elimination of truck rolls to check meters.

Some dirty math:

• Install cost of utility smart meter (all in) — say $100
• Hourly wage for meter checker — $25/hr
• Minutes needed to check each meter — 3 minutes on average, so thats $1.25/check.
• 12 checks per year, so $15 saved/meter/yr
• Payback = 6 yrs and 8 months, which fits in pretty darn well with utility asset financing.

So — at the risk of being stupidly outspoken — here’s what I’m thinking:

When the value of a deployed data technology can be quantified in the payback period on the physical asset — that’s IoT.

When a deployed data technology is valued by the automated services it creates -that’s industrial machine learning.

Silver Springs and Itron probably want to become utility machine learning companies. They could, but they won’t — and for a couple good reasons:

1. They’re already good at what they do, and don’t really need to.
2. The sampling rate for most utility smart meters is < 1 Hz, and for utility operations that’s not granular enough to provide value from machine learning today.

PingThings and High Frequency Time Series Databases

Utility synchrophasors, on the other hand, are capable of sample rates greater than 1000 Hz. Thats a data deluge when scaled to thousands of sensors across a utility network, so much so that just acquiring the data breaks current state of the art providers to utility markets.

So, the PingThings dudes out of Berkeley built one of the world’s best databases optimized to ingest high frequency time series data inputs.

And while they entered their pilots hoping for moons and stars — as most startups do — what they discovered was an extremely strong demand for the voltage sag detection capabilities of their platform.

Pingthings are the only company in the world offering a searchable, explorable database capable of capturing and tracking millisecond-timescale, transient events — like a voltage sag — over thousands of sensors and years of operation.

Hmmm — thats a mouthful. And at first there wasn’t a lot of good understanding or agreement about why this new capability is valuable.

But now that the capability exists, the probability that just this single capability becomes a reporting requirement for all utilities is greater than zero, and it is already under review by regulators around the world.

There’s your gateway, and there’s no better stickiness than being told to use a technology.

But what’s the growth opportunity? . According to one of their utility beta customers, voltage sag represents less than 1% of the capabilities of the Pingthings platform, and utilities currently know only about 10% of what they believe the full value will be.

And that’s an exciting point to build from.

KISS and “The Big Deal”

The future of electricity operation combines high precision, automated grid management with high levels of both distributed generation and electric vehicles working in economic synchrony.

The technology to do all this exists today, and the process of implementing it at scale is a function of understanding, coordination, investment, and will. But the amount of operational effort to make this fully functioning dream is massive, with a design that’s currently undefined.

As I noodle on this with the PingThings team, I keep thinking back to one of my earliest mentors, Doug Wert . Heaven got a doozy with old Doug, and when he wasn’t being an outspoken, brash, over-zealous project finance savant, he was browbeating a young Andy to ignore complexities: “keep it simple, stupid!”

KISS — an early lesson in the value of simplicity, and one of the hardest things for technology-bent early stage startups to master.

PingThings has mastered simplicity in their first product offering, one that plays a small but critical role in the complicated, data-driven, multifaceted future of electricity generation and delivery.

More importantly, PingThings is not just knocking on the door, they’ve already walked in.

That’s incredibly valuable, and an exciting first step towards more reliable, safe, distributed, sustainable networks — electric and otherwise.

Exploring Oceans of Firehose Data was originally published in Comet Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.

Chances are, unless you’re already in the semiconductor industry, you probably haven’t heard of the term “RISC-V” or know what the letters stand for. Chances are, if you’re like me and blindly believe in the power of phonetics, you might not even know how to pronounce RISC-V.

When I first met with Professor Krste Asanovic, the head of RISC-V.org, on Berkeley’s glorious grounds, I kicked off by saying something highly eloquent. Something like, “So, Mr. Scientist. Tell me what you know about this ‘risk vee’ stuff.”

“Uhh, it’s pronounced ‘five.’” Eloquence everywhere.

“What’s that? Five what?”

“Risk five. RISC-V.”

I didn’t talk much thereafter, yet Professor Asanovic was still kind enough to hold my hand and walk me through an ocean of questions.

What I then thought was going to be a straightforward research project into a new microprocessor technology ended up opening a can of worms (and a couple Coursera classes for computer architecture) that convinced me we’re on the edge of a new wave of silicon innovation, and witness to the kind of semiconductor companies we haven’t seen for a while here in Silicon Valley. Since the good old days, when band gaps meant something . Putting the Silicon back in Silicon Valley.

Simply put, RISC-V is an open-source instruction set architecture (specifically, Reduced Instruction Set Computing, hence RISC), defining the “rules” for software programs to communicate with the physical components of microprocessors like memory and logic units. It’s a major threat for incumbent semiconductor companies that work hard to market incremental gains in their proprietary technology while defending it from competitors and innovation through vast IP portfolios.

While an open source instruction set would be a boon for semiconductor designers working on new compute applications like AR, VR, and machine learning, this kind of implementation in the industry requires a galaxy of support and fabrication services, a unified effort across engineering and design capabilities not just to build functional RISC-V processor but to ramp and integrate them into servers and devices.

The RISC-V concept launched out of UC Berkeley way, way back in 2010. Take a look at the ecosystem built around the RISC-V community that has risen from nothing in less than 8 years:

Though many of these companies are startups, you’ll see a few big names on there who are starting to understand RISC-V before a startup beats them, and preparing for an increasing reality of an open-sourced, highly flexible ISA (instruction set architecture). Western Digital, for example, has taken a bold lead in the future of chipsets from diverse manufacturers and applications, all based on the RISC-V instruction set architecture.

The RISC-V Movement

RISC-V in itself isn’t a marketable technology. By design, it’s a platform that allows anyone — researchers, companies big and small — to build chipsets with an unlimited range of applications on top of a standardized, open-source instruction set.

Now, ISA’s are a unique class of engineering challenge — they’re not necessarily difficult to build, but they are very difficult to build right and in a way that’s both scalable and free from errors. After all, these are the things that manage information flow between the make-believe fantasyland of software and the brutally physical hardware realm, telling gajillions of 1’s and 0’s where to go and every second between the two demands careful housecleaning. CPU history is filled with examples of ISA’s experiencing specific, unforeseen failure modes after mass product deployment. And because the instruction set defines how software communicates with the underlying hardware, patches to the ISA in the field aren’t possible (ergo, hardware is hard).

The challenge of building highly functional ISA’s is therefore a valuable capability, bordering on an art form. And the leading artists — Intel for datacenters, ARM for devices, NVIDIA for GPUs, etc. — have become very big companies in large part because of their ability to both build excellent, extremely reliable ISAs and (most importantly) control what anyone does with those ISAs.

Its this second issue — design control — that’s particularly interesting. When someone wants to build a new class of microprocessor or processor unit, they either have to build out their own ISA or layer their new applications on top of an existing ISA.

The value of RISC-V is for innovators to layer new compute applications on top of a highly capable, reliable, and open instruction set.

RISC-V is the result of the unified effort to design an ISA that is universally accessible and available to use for many devices and applications. Now, anyone can use the RISC-V open-source ISA as a base level to design a chip and integrate it into a computer, decentralizing the authority from a few dominant companies and putting it into the hands of entrepreneurs.

In essence, RISC-V has created a new platform for semiconductor innovation. It’s leveled the playing field for building new chipsets with novel applications — new, potentially risky applications that may not have clear market pull yet but that can launch at the “beta” level without investing millions of dollars to accessing ISA IP or design services.

For entrepreneurs, it’s the sticky matter of identifying which market demands are best satisfied with an open source ISA — where they can be competitive to traditional processor companies — and defining the path for innovative new chip designs that will carry our computing capabilities forward.

So, should you care?

If you’re in the semiconductors industry, absolutely. And if you’re an electronic device manufacturer, absolutely. And if you buy smartphones or laptops… it’s ok if you don’t, but you might see some strange new logos popping up.

If it’s successful in its mission, RISC-V will increase the rate of compute innovation by lowering the barriers to entry for high performance, reliable, fundamental compute technology.

Consumers will see faster, better devices with wild and useful applications — in machine learning, virtual reality, blockchain, who knows!

But behind the scenes, they can thank RISC-V.

The Next Silicon Innovation Era was originally published in Comet Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.

Comet’s Assessing China program for Industry 4.0 startups had a simple goal — bring the top AI/robotics startups in a specific industry to China to meet with leading companies and de-risk go-to-market partnerships for both sides.

That’s bold in the best of times, and downright maniacal in today’s political climate (thank goodness they didn’t slap a tariff on airplane pillows).

But innovation is a universal language — the push and search for the better, the new, is a business reality around the world — and the timing is now for data-driven automation of massive industrial operations.

It’s already happening.

China is the focal point of global manufacturing, so it made sense to have our first Assessing China trip focused on industrial IoT. We selected five top venture-backed startups working on “Industry 4.0” to explore go-to-market partnerships in China based on product readiness, technology strength, and market fit. Those startups were:

• Arch Systems (Bay Area)
• Augury (Israel/New York)
• Canvas Technologies (Boulder, CO)
• PsiKick (Bay Area)
• Ready Robotics (Columbus, OH).

While in China, the teams met with a diverse group of manufacturing leaders including BYD, Lenovo, BGI, and Hytera. We identified current stakeholder pain points and explored the broad range of China manufacturing setups. Further, the teams had the unique opportunity to engage with strategic leadership from Saint Gobain — one of the largest industrial producers in the world with a large operational footprint in China — as well as leading investors in China looking for unique solutions in the startup ecosystem.

And it worked — partnerships are being formed, and Industry 4.0 technology, driven by startups, is being deployed in China. Ground truths discovered from these visits were both unexpected and enlightening, and focused on two main points:

1. Mass manufacturing has a people problem.
2. Smart Manufacturing means Flexible Manufacturing — and this is really hard to do.

From the Founders: What worked, What didn’t, and What we learned

This was Comet’s first Assessing China effort. A prototype. And like all prototypes, we had some preconceptions of the intention, goals, and outcomes of such a program.

As with any prototype launch, however, what we discovered through our test effort was just as valuable as all our combined knowledge guiding our predictions.

For the startups, the greatest value of Assessing China was in learning: about the market, about other startups working in their space, and about themselves.

Learnings about the Market

One of the most exciting findings was from Gal Shaul, the founder of Augury, noting the “overwhelming response to automation and the interest of industry to adopt”.

In my previous experiences as a founder, interactions with potential business partners in China felt like a one-way street. I’d pitch, describe my technology and product and how it could help their business, and they’d listen, usually through an interpreter. The end of the first meeting was usually a general politeness and an agreement to continue the dialogue.

Not so with our meetings on this trip. One of the early manufacturer visits with a larger paper manufacturer began in this vein, but by the end of the site visit they were describing key pain points — quality control — and how AI companies could help their QA processes.

Another manufacturer, a mid-size injection molding operation in Shenzhen, politely listened to the companies discuss their technology and then directly said: “We don’t have a data problem. We have a people problem.”

Wow! Their head of manufacturing operations went on to discuss in detail his staffing issues and key pain points of maintaining high-quality work in the face of increasing automation. His workers get bored, and he’s looking for automated solutions to solve their boredom problems.

These kinds of honest, direct insights increased with the size and sophistication of the manufacturer. Personally, the most compelling moment of the trip was at the manufacturing campus of BYD: we got the tour, then an audience with management in a conference room, on the floor above their massive injection molding tooling operations. BYD’s team listened to the companies discuss their products and technologies in turn.

Then, company by company, BYD’s head of manufacturing operations individually and directly reviewed the startups in turn: we don’t need this; your solution is interesting; we’re already doing that; we’d like to pilot this now.

As a team, we’d not experienced this kind of direct, honest feedback from groups in China. The key lesson is that there’s a growing interest in China for innovation and acceptance of working with early stage companies. This is driven no doubt in part by an increasingly active innovation scene in China itself, where VC activity is increasing at a very steep rate to become one of the largest innovation centers.

Hytera, a leading provider of professional radio communications and partner with Siemens for intelligent manufacturing transformation, put it like this:

“We have explored a complete set of intelligent manufacturing development plans. This exchange event allowed us to see a lot of good overseas startup teams, learned about the latest development trend of ‘Smart Manufacturing + AI’ technology in the world, and also enabled overseas startup teams to have a deeper understanding of Chinese manufacturing.”

And from Andrew Scheuermann, co-founder and CEO of Arch Systems:

“The Assessing China program exceeded my expectations in the relationships I was able to build with many of the manufacturers as potential future customers, as well as the fellow participants.”

Right on, Dr. Scheuermann. Right on.

An Unintended Founder Network

Three of the companies in this first Assessing China program — Arch, Augury, and PsiKick — leverage sensor data to enable and optimize manufacturing automation. The other two — Ready and Canvas — are technologies for industrial robotics.

To be honest, we didn’t know what to expect when throwing these teams together in a place like China. Founders can be sensitive to potential competitors in their personal space. While we anticipated some tension, an all out street brawl in the heart of Guangdong was not out of the question. We prepared for that. Mentally, at least.

But a funny thing happened on the way to that first manufacturer meeting: traffic. Hours of it in a van. With nothing to do but talk.

Andrew from Arch said, “We built friendships and learned from fellow industrial entrepreneurs that I might not otherwise have met and spent a week with. It accelerated my knowledge about the bigger Industry 4.0 picture as well as gave me perspective and wisdom from others’ journeys.”

Indeed, by the second day, the teams had refined their company introductions to show how their technologies are necessarily, mutually collaborative for the continued advancement of manufacturing digitization and automation.

One of my favorite examples of this came during conversations with MGI, a leading manufacturer of genetic sequencing machines. They want to reduce the number of robotic arms required to automate their manufacturing operations. Together we discovered that Ready Robotics on a Canvas AGV provided a unique and creative solution.

As a founder in the Valley, one of the most valuable things I found going through accelerators was building a founder network and leveraging the mind share to strengthen the whole community and avoid repeating mistakes.

In talking with Benjamin Gibbs from Ready Robotics, it was those hours on the bus — rare downtime in a founder lifecycle, and even more rare to be in the company of other founders working in a similar market, and at a similar stage — that were perhaps the most valuable during the week. It was an opportunity to learn, to research, and to better understand how founders can collaborate as they grow.

Direct Feedback and Fast Reflections

Never believe your own hype. No quote necessary — this is a universal truth for startup founders. Yes, there is a large part of salesmanship, marketing, and PR involved with startups, especially if they’re VC-backed. The pressure is real and the competition is fierce.

With startups focused on B2B with industrial partners, the need to demonstrate real world values and KPIs is intense. There is little to no room for BS when you’re talking about integrating a new technology into an assembly line or industrial setting. If you pilot ahead of product readiness, you may get a short term investment but a long term kick in the ass.

The direct, repeat feedback forced the startups to hone in on their unique value propositions. This was extremely important. Overall, we discovered that while China is indeed open to innovation, and companies are increasingly willing to engage with early stage startups, the burden of proof is still on the startups.

China can be a very compelling market, but startups should only consider going into China when their products and services are clearly defined and ready to roll. This was one of the few preconceptions that held through the experience, and the results are telling.

“This has been the most eye-opening and business progressive trip I’ve ever participated in and I would recommend any startup CEO/CTO/COO/CRO to join if they are thinking of the Chinese market.”

— Nils Alstadt, VP of Customer Experience at Canvas

Views from the (Comet) Tail: Our Thoughts as an AI-focused VC

Our first meetings in Shenzhen were on July 9th — a few days after $64B in tariffs were levied against China goods and a day before the threat of an additional $200B. Interesting times, and definitely top of mind for many on the expedition.

However, as is always the case, it quickly became clear that innovation and entrepreneurship are universal languages of global interest. New technology is always compelling and there is a strong desire to make partnerships work in an increasingly globalized Earth.

In addition, there is a growing openness from Chinese businesses in discussing key industry pain points and areas of technology interest. This is exciting stuff, reflecting an increasing awareness and appetite for new technologies to accelerate into traditional industries — in many cases at a significantly faster pace than US and European counterparts. While there are certainly complexities, China is certainly open for business and is pushing technology adoption. Just try to buy a hamburger with cash at a McDonalds in Beijing (spoiler: you can’t).

In the manufacturing industry specifically, while there is a lot of talk around leveraging AI tools to push manufacturing automation to the limit, there is a strong understanding in China of the limitations of the technology and where, how, and why these technologies ought to fit into today’s operations and the manufacturer of the future.

Hytera in particular is pushing the envelope of manufacturing automation, leveraging robotic arms and line optimization to a degree previously unseen by Comet Labs or the visiting startups. Their partnership with Siemens reflects this, and their combined effort is remarkable.

Increased automation in manufacturing settings decreases the workload on individuals. One of the repercussions of this is the tools available for failure analysis, both in individual parts inspection and in machine O&M, are increasingly boring. Even for highly unskilled labor — US or China, it makes no difference — these QA and FA tools make the job boring to the point of distraction. Coupled with the fluid job market in China manufacturing, where employee turnover approached 70% annually, it’s becoming increasingly difficult to just hire people, let alone keep them engaged in the job long enough to have an economic advantage for the factory owner.

This is the driving motivation behind the analytics and failure prediction companies like Augury, Arch, and PsiKick. Yes, there is value in having an automated system do a better job than people, but it’s a marginal value. The problem is that manufacturers today cannot find and maintain people to even do the job. Deployed sensor networks, integrations with Manufacturing Execution Systems, and layered machine learning services are universally viewed as a necessity in an increasingly automated future.

Flexibility is King

Mass production is build around the concept of an assembly line — a large investment in rigid machines and linear workforces that can then leverage economies of scale to output a given product.

Two trends are shifting the calculus of the assembly line:

1. Demand for customization
2. The compressing timeline of product innovation cycles.

Rather than automation offering potential operational savings from reduced labor costs, the value of automation and data-driven services is in building an increasingly modular, flexible manufacturing setting that can quickly adjust to new products without sacrificing operational cost.

Here, one of the most compelling potential solutions was a hybrid between Canvas Technologies — the leader in autonomous mobility systems in unstructured indoor environments — and Ready Robotics, the universal OS for robotic arms. By offering a mobile, highly programmable robotic arm, manufacturers can potentially have a single arm operating at multiple points on an assembly line, and one that can be reconfigured for different tasks with minimal downtime. Comet Labs and the manufacturing partners had not seen a solution like this in the startup world or elsewhere, and the potential combination is a unique, collaborative finding of the Assessing China effort.

Technologies that enable manufacturing flexibility is increasingly important when we consider the trend toward a single machine vendor for a complete assembly line. The value here is that all machines would speak a single language (there is no standardization for industrial machinery across vendors), the data is centralized, and operations can be optimized across the line. The downside is the vendor, today and in the future, maintains control of the data and offers that optimization as a service to the manufacturer.

At risk of upsetting some industrial stakeholders, this concept of centralized data limits the adoption and scaling of fully flexible, autonomous manufacturing systems.

Disruption is at the heart of the startup experience. The major disruption for the startups on this Assessing China trip was that they unlock the value of data for manufacturing owners, enabling deeper levels of automation and flexibility without sacrificing time or operational margin.

Onward!

“Ignore the news and focus on doing business with people you trust.”

- Gal Shaul, co-founder and CTO of Augury

I really like Gal’s style. With startups, there’s always dozens, hundreds of reasons not to try something new, and only one or two reasons to go for it.

There’s so much noise. And at a time of political tension between the two biggest economies in the world, with AI as one of the foci of that tension, Comet decided it would be helpful to bring the top AI startups in industrial automation over the Pacific to meet with some of the manufacturers leading China’s efforts toward Industry “4.0”. There were plenty of risks — the wording was buzzy, the timing was questionable, and the concept had definitely been tried before.

But something interesting happened once we actually got to Shenzhen. China opened up, it listened, we listened, and we all saw a step taken to introduce and grow new US technologies to the largest scales in Chinese manufacturing.

I personally would not have believed it possible before we went. I do now, we’re excited to help these startups continued to accelerate into Chinese markets, and we’re already working to start the next group journey of Assessing China. Who knows what opportunities it will create!

To stay up to date on future Assessing China programs from Comet Labs, sign up on our website or send us a note at hi@cometlabs.io.

Ground Truth in China: The Push Toward Flexible Manufacturing was originally published in Comet Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.

China is a bit of a black box when it comes to entering as a foreign business, and as a founder and investor I’ve seen a couple companies — specifically startups — get messed with.

In China, no one messes with Legend.

It’s only logical to choose manufacturing as our first industry-focused Assessing China program. China is well known as the global center for large-scale manufacturing, and its growing investment and adoption of IoT and AI technologies positions China not as testing ground but early adopter of (I suppose we’re calling) Industry 4.0.

First, a little about Legend.

Legend Holdings is one of China’s largest and most respected conglomerates. Famous for owning the controlling stake in Lenovo, Legend is an industrial leader in China owning 40+ companies in several industries including agriculture, healthcare, manufacturing, and financial services.

To lead investments in frontier technology startups, the company created Legend Star, one of China’s oldest and largest early-stage VC funds. Legend Star has invested in over 200 companies. To support growth stage opportunities and market entry partnerships, Legend created Legend Capital and Hony Capital — since 2016, Legend has been the go to market partner for WeWork in China.

Comet Labs is the US investment arm of Legend Star.

China is the global center for manufacturing operations — no debate about it.

To help scale manufacturing tech faster, we — in conjunction with our partners Legend Holdings and Lenovo — are bringing leading “Industry 4.0” teams to China to sit down at with the country’s manufacturing leaders, define strategic partnerships, and accelerate deployment of manufacturing startup technologies at the largest scale.

According to Wikipedia, ‘twas the Germans who coined the phrase Industrie 4.0 in 2011. Looks like industrial progression goes something like this:

Honestly, I didn’t know there was a third industrial revolution. And I don’t really know what “cyber physical systems” means. It sounds super menacing though.

I do know there is massive potential to leverage computer vision, robotic optimizations, predictive analytics, and other machine learning tools to automate manufacturing operations and make them more efficient, sustainable, and economic.

This is Industry 4.0.

China is a huge market, it’s growing rapidly, and the attractive pull for new technology and innovation is strong. For a few industries — manufacturing, healthcare, autonomous and electric vehicles — China’s national strategic push is driving technology adoption and innovation faster than anywhere else in the world.

Simultaneously, time is the most precious resource for startup founders. And while China may be the greatest market opportunity for many of today’s AI companies, the time and focus functions of a China go-to-market strategy demand entry be pushed back, sometimes indefinitely.

Hence the motivation for Comet’s new Assessing China program.

Unlike previous “Discover China” tours — which I participated in as a founder, our head of business development Taylor Stewart worked on building at ReadWrite Labs, and Lucas Wang on our team built from the ground up with HW Trek — Comet’s Assessing China program is designed to:

1. Identify strategic market entry partnerships between US startups and corporate leaders in China;

2. Assist and support leading startups to execute partnerships with terms that enable them to succeed and grow fast;

3. Leverage Legend as a strategic partner to support the efforts us US technologies and startups.

Assessing China is designed to be as impactful and as efficient as possible, both for startups and manufacturing leaders in China:

• Small, focused, high quality cohorts — companies can send us their information here. Comet will identify the TOP FOUR companies leading the transition to Industry 4.0.
• Preliminary information sharing — Before the trip, Comet will prepare and share company and technology briefs with participating manufacturing partners in China. Conversely, Comet will share information on the manufacturers to cohort companies. The table is set for introductions, and the intentions of the visit clearly outlined.
• One week, three cities, six partners — Comet covers travel costs for cohort companies from the US to Shenzhen, Guangzhou, and Beijing to meet with leading automotive, electronics, and apparel manufacturers.These are focused meetings, to grow rapport between your company and theirs, and to focus conversations on strategic alignments and strategies.
• Deal Execution — Comet and Legend will support the cohort companies in their communications and deal makings with identified manufacturing partners.

The goal of Assessing China is to have the leading US manufacturing startups come back to the US with new personal relationships with corporate leaders, and a clear path to partnerships that will elevate both technology companies and the manufacturing leaders looking to deploy new, leading technologies.

Ready to apply?

Assessing China is a best fit for mid to late stage startups ready to deploy leading technology solutions to the largest manufacturers in the world. Startups should either have a product that you’re already testing here in the US and need a bigger market, or you have several clients (nice!) and you’re hunting for new, large customers.

You don’t need to know anything about or have had experience in China to be a good applicant. You should just be interested in the opportunity!

More information on the program is available at our website, where you can fill out the quick form about your company. Then we’ll be in touch.

Questions? Shoot us a note: hi@cometlabs.io

Unlocking Industry 4.0 with China was originally published in Comet Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.

Artificial intelligence — as a technology, not a concept — is entering its design phase.

In a past life, I had the opportunity to learn an important lesson about design thinking when I was founder of Stower Energy. Equity makes magic sometimes, and in the instance of Martin Schnitzer it got me several months working elbow to elbow with him in a small studio in the Excelsior district of San Francisco for our startup. Martin spent years at IDEO as a designer, and while we had a shared passion of sustainable energy systems, what’s really great about Martin is that he’s expert is what he does.

And what Martin is expert at is getting to the damn point.

In those whimsical months, Martin worked his design magic and helped Stower cut through the weeds of innovation, digging deeper into the essence of what we were trying to accomplish. He honed our focus 100% on what we wanted to build and why, until we finally got to the nut — the actual value of a thing we were building.

It saved us valuable time and energy, and I’ve learned to see the value in that kind of thinking — design thinking — in founders.

This is the appreciation that led us to become one of the largest investors in Arch Systems. Arch has built a breakthrough method to extract data from any machine and transform it into insights. Most importantly, they’re bringing the capabilities of AI to any system, regardless of how “smart” the machines already are.

Today, the AI/ML/Robotics technology wave is rolling past awe at technological capability to actually delivering products and services with value in the world. Some of the best AI technology today is in ad delivery and voice assistants — transformative pieces of technology in the digital world. But it’s the pending implementation of an action based on a desired goal-state in the physical world that is truly at the heart of Comet’s thesis in Intelligent Machines.

As access to a technology broadens, different minds start building different things. When a unique combination of personal experience and technology comes into reality… BAM! You have an evolutionary step. Like fruit flies. Like Uber.

For the manufacturing sector, we’re seeing one of those evolutionary moments right now. Let me explain.

In a manufacturing company today, decisions are typically made based on some analysis of information available in order to discover a bottleneck that is limiting capital creation. There are currently solutions available to automate all of these steps in some capacity which largely fueled the “big data” revolution.

At a high level, this enables a shift from a more detailed understanding of our world while automating what we can, to an experience where technology helps us define higher level goals and machines help us reach them.

Or in the words of Tom Soderstrom at JPL, from industrial automation to artificial intelligence. IA → AI

Based on this evolutionary trend, a logical customer target for applied AI/ML/robotics are owners of large, diverse machine fleets. In industrial settings — factories, utilities — the goals are fairly simple:

1. Create new products that provide a competitive advantage vs. the current universe (creation).

2. Make production of existing products more efficient and safe (optimization).

Collect data from machines, categorize it, run it through a trained NN or twelve, throw an actuator on a button, and presto! We’re living in the future.

There’s a couple problems with this.

The biggest problem is that >99% of industrial machines in the world today are not smart or connected. Even if they were, getting fleets of diverse machines to all talk and coordinate with one another via a central control — without any kind of economic value — negates rapid adoption of technologies beyond meaningless pilots.

Instead, industrial leaders budget for technology adoption based on the replacement cycle for their machine fleets which, in many industries (apparel, utilities, metalworking, etc), is measured in decades.

What’s really needed to complete the path to an insight is what we call Stage Zero.

Stage Zero is called various things — industrial IoT being one — and its where this evolution begins for today’s industrial infrastructure. It’s taken me a while to appreciate just how important State Zero is, as our minds can’t grasp it until we see all the way to Stage Five and know where it ends. And I don’t think I could have gotten here without Martin’s continued design whippings in our quest for unravelling the true value of a thing. But when we got there, we realized just how much value is actually in a machine and how many insights can be extracted from its essence.

This is where Arch Systems comes in.

Their technology is fascinating; they can measure everything from shock, vibration, current, and voltage decoding into uptime/ downtime, performance, and other predictive analytics. What they do is right on, but it’s how they do it that’s revolutionary.

They’ve come up with a way to automate, from the design phase, the deployment of industrial automation to “arch” over the full evolution of an insight. Their technology lets them go to a manufacturing line, with tons of different machines, and make sense of them all to move beyond Stage 0 and chain machines together for an evolutionary system, to take existing industrial infrastructure through prototype to machine intelligence over 10x faster and cheaper than any previous technology.

Arch Systems is the best in the world at automating industrial machine fleets.

In the coming age of AI, Arch’s process means more people can see the benefit faster, and we couldn’t be more excited for that future.

I used to tell a very lame joke when I founded an energy IoT company:

What did the air conditioner say to the sewing machine?

Answer: Watts up?

But I think there’s a new, better answer: Ask the team at Arch.

The Startup Designing the Evolution of Industrial Insight was originally published in Comet Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.

Today’s automation tools are boring.

It doesn’t matter what the application is — playing Go, picking fruit, stacking boxes, annotating dogs on a highway — the state of the art in artificial intelligence is a bit dull and Siri’s jokes suck.

As it should be! It’s taken us a long time and technological innovation to get to deep neural nets that tell us what exactly we’re holding in our hand and a machine to respond as best as it can to our clumsy, wayward Q&A’s:

“Hey Siri, play me some EDM.”

“You got it!”

“Google — what’s the weather?”

“In San Francisco it is 68 degrees and foggy. Again. Like yesterday. Like tomorrow.”

Form follows function, and up until now our focus on applied AI technologies has been on execution of the technology for industry-specific applications: pick an apple, predict a machine failure to avoid downtime, deliver medical tests from home to hospital, answer our questions. And if the technology doesn’t produce a real world value, Comet Labs probably won’t invest.

But we’re always confronted with some pretty fundamental questions as we walk our investment path, like: What’s the value of this application? Who will be the ultimate recipient of this value? Yes, there is massive value in the current state of the art in machine learning and AI, but the capabilities are effectively boxed into “industrial automation.” IA.

To steal a concept from Tom Soderstrom at NASA’s Jet Propulsion Lab, we believe the next interesting shift in this universe of machine intelligence will be from IA to AI. Automation to intelligence — real AI. One that doesn’t just answer questions, but helps the ultimate recipient of the value accomplish an important goal.

That takes us to Akin and the company’s founders, Liesl and Travis.

Akin are building tools and technologies that solve human problems with machine intelligence. Based on neuroeconomic principles, abstract reasoning, autonomous exploration…Wow, where’d all that buzz come from? Usually when we see summaries like this we get all introspective and stuff — what are these guys actually trying to build?

But with Liesl, she’s already built it. She sold her last AI company, Cognea, to IBM which then became a critical part of Watson.

On that statement alone, Akin is a force to be reckoned with. However, it’s truly rare to find a founding team whose passions, experiences, and expertise are individually unique while at the same time radically complementary to the company they’re building. Coupled with the “real world” partners circling wagons around Akin’s technology and vision, we believe Akin is walking a path to massive deployment of novel, goal oriented, Beneficial artificial intelligence.

Yes, that’s Beneficial with a capital B — Akin is a Public Benefit Corporation. As a former “social entrepreneur” myself, when they told me about their legal status I initially coughed, coffee misting from my nose and mouth, dribbling down my chin.

“What in the hell are you maniacs peddling?” I screamed. “A B Corp? Are you serious? A B Corp?” This was in Comet’s offices, high in the fog at 3rd and Market. Windows rattled. Our robot-themed wallpaper blistered from the hellish vibrations. Interns scattered. “A B Corp?!?”

But their answer surprised me.

A key piece of what’s missing from today’s technologies and interfaces is an enhanced human-machine relationship and understanding. Its a design thinking effort, coupled with remaining scientific and engineering challenges to current technology architectures. How do we shift ML technologies from solving problems to helping people achieve goals, given that the human state at any given time is in flux? How do we make AI tools more engaging, so we learn from machines and machines learn from us over time?

A few years back I took my dogs, Carne and Queso, to a training class. They’re awesome dogs, but they’re rough around the edges. They eat what they shouldn’t and bury my socks in the filthiest corners of our backyard. Carne especially. I honestly can’t keep a pair of socks for more than a month before they disappear into the mud.

As anyone who’s taken their dogs to a training class, the challenge really isn’t to train the dog to do what we want them to do. The thing is to train ourselves to communicate with our dogs, to understand how to communicate with our dogs, and learn how to work together to get them to sit and stop hiding our socks. Its mutual; our pets help train ourselves to be able to train our pets.

A dog that only does what we want it to do is boring. The best dogs are the ones that we learn and grow with, that understand what we’re after when we say “don’t eat that” and that’s spirited enough to wag their tail every time we come home from work.

We believe AI can evolve in the same way.

Through better understanding and engagement with the technology, we can create a tool that is truly valuable for us, that helps people define and accomplish their goals in an increasingly complex, fast moving world.

And we believe Akin are the best equipped in the world to build it.

To keep up with our investments, hit the follow buttons below and subscribe to the Comet Labs newsletter for a bi-weekly digest of our favorite AI and robotics stories!

Our First Investment in Artificial General Intelligence was originally published in Comet Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.

Dear Founders,

A couple weeks back I was talking about YC, demo days, and how it made us think through our beliefs at Comet Labs. I also brought up the concept of SoftBank’s Vision Fund as a Black Hole. It got my mind whizzing a bit and going off on too many tangents, all hypothetical with little basis on quantum anything. At the end of the day, the Vision Fund is having an impact on early-stage investing and we need to understand it.

Now, I’ve re-written this damn thing three times and ultimately landed here as a simple letter to you, the founders of early stage startups, and perhaps my future self when/if I scamper back into that odd existence of the startup damned.

As founders of venture-backed companies, there are really only two important things to keep in mind:

1. The burden of proof is on you. It’s your job to to prove your vision is valuable and to make it successful. The Valley is here to help, but no one else is going to do it for you.
2. You get compensated when you’re proven correct with an exit… NOT at an investment round.

Whew! Well, that wasn’t so hard now, was it?

I bring this up in the context of SoftBank’s Vision Fund, and why I’m having such a hard time getting it. I had to whiteboard the heck out this thing — in solitary confinement, then with my cellmate and Comet brand director Derek Embry — and we came up with a couple interesting concepts that have helped me get a grip on something massive and unprecedented in the venture landscape: the $100M+ early-stage raise.

My conclusion? Ignore it.

I know, I know — it’s nigh impossible to ignore. Big money is super sexy.

The black hole analogy for the Vision Fund is good because its internal mass is so breathtakingly huge. I mean, look at their portfolio:

It looks like a gravitational well, or half a tornado, with Brain Corp at the bottom with their “measly” $110M (or, as Recode puts it, $0.11B).

Ignore it.

Here’s why:

1. $100M at early-stage limits your ability to pivot if you learn something new about your customer.
2. $100M at early-stage limits your options for exit.
3. $100M at early-stage isn’t defensibility.

Raise your rounds, make your mistakes, keep on chipping away at building amazing things. Don’t raise a lot of money or post high valuations with the aim of attracting the fastest path to success. Don’t do it.

Life is short, so let’s stop getting distracted and get back to work. Focus, focus, focus!

Myself included, and this is the last time I’m going to think about the Vision Fund… until they come to me with $100+M. Then maybe I’ll think about it again.

Yours in the grind,

Andrew

PS — if you want to know why I think founders should ignore this new wave of ultra large funding rounds, read on… if you dare.

One Fund, One Vote: The Art of Managing Authoritarian Voices

$100M from one investor at early-stage limits your ability to pivot if you learn something new about your customer.

Raising money from a venture fund is kind of like votes in an election. It’s not that investors know a given company will win, they’re just casting ballots with the belief that the company will win.

Sure, we early stage VC weirdos cast votes in lots of different elections hoping “1 out of X” will be a big winner, but at the end of the day there are thousands of ways startups fail regardless of how much money they raise.

In our current universe, a company can now raise $100M+ from a single investor at an early stage in it’s evolution.

With a minimum check size of $100M, the voting power of the Vision Fund is extremely high. That much cash, especially in early-stage companies that have not yet proven their business model, gives the recipient a massive competitive advantage over all others working in the same space.

But $100+M comes with a price, namely a company valuation high enough to warrant such an investment.

This is fine because the basis of startup valuations is more art (or hustle) than science. But once you establish that valuation, its stuck to your company like spit from a llama.

So let’s say you raise a grip of cash at a high valuation. It’s now up to the team — founders AND investors — to prove your company is worth more than the new, massive valuation and start your glorious march toward startup success.

But as it invariably will the sky falls as some fatal flaw presents itself. An unknown unknown. One of the thousand possible pitfalls. Zounds.

Then it’s time for a change and by the luck of all leprechauns, this happens to be the very best thing in the world startups are really good at. If a startup is the only one on the world doing its thing, pivot away and take all the time you need.

But startups rarely operate in a vacuum, and for most startups there’s at least ten others working on a similar concept with each at similar phases in their development and understanding of the market. The line between success and failure is blurry and a combination of multiple factors. Cash is definitely one of these factors, but not the only one.

Of the following, which company will pivot faster?

• The company that took a $100M investment — and the requisite board seat that ought to entail; or
• A competitor that raised less money, has less runway, and is still fighting for their very lives on a cash flow basis.

Our thinking is the latter. This highlights one of the flaws of large check sizes for early stage companies:

The Vision Fund is shifting Series D to Series A, reducing founder flexibility by eliminating gates founders should pass through to access greater capital.

So the advantage for the startup taking big money isn’t speed. It’s the opposite: time. They have more resources, a longer runway, and an investor that believes enough in them to invest $100+M to help weather major storms during a startup odyssey of epic proportions.

But all that time gives startups a false sense of confidence that their solution is the correct one before anything is actually demonstrated.

What they lose is the urgency to get stuff done and cut the BS; to jump on any opportunity or signal of a customer pain point, because the company’s life is at stake.

Money often makes people do stupid things. Big money and high valuations make people complacent. That’s the reality for most of us mortals, and it’s everything we’re taught not to be as early stage entrepreneurs.

Ultimately, if the Vision Fund deploys $100M minimum checks in early stage companies, it must police itself against being too rigorous on how that money is deployed and allow founders the flexibility to change their strategy…even risking all in the process. If they don’t, investors risk bloating a solution to a problem that doesn’t exist, while sustaining something the market doesn’t want. And if that were the way, we’d all still be reading magazines and buying our duds at JCPenney (I do both, but am outside Amazon’s core demographic). Disruption is all about doing things better than the other guy, regardless of how big the other guy is.

Sometimes You Eat the Bar, and Sometimes the Bar Eats You

$100M from one investor at early-stage limits your options for exit.

One simple thought experiment we ran through at Comet HQ was to play “What would you do?” scenarios. It’s a fun, daydreamy kind of thing.

You’re walking down Market Street, and some geeks in a suit offer you $100M (or $0.1B). Would you take it?

Kneejerk? Heck yes I’d take it. That’s a ton of cash to play with!

We could:

1. Grow large, massively inefficient sales teams and make big money mistakes bringing new concepts to market; or
2. Rapidly flip the company in an acquisition or IPO, where we’ll face the ever increasing skepticism of a world outside Silicon Valley; or
3. Acquire other companies to diversify our offering.

But then I realized something. Point #2 raises the second ugly head of this kind of massive investment in early stage companies:

High valuations end up limiting exit opportunities, effectively forcing startups toward an IPO.

Hey, I went to Stanford. I did my incubators, my d.school. I know things. And what I know is that there are way more acquirers who can afford to spend $20M buying a company than $200M.

A lot has been written about what defines a good exit, and how focusing on multi billion dollar paydays is a bit silly. From a statistical standpoint, founders taking on a high valuation limit themselves to an ever smaller pool of potential acquirers. At the same time, they put themselves at risk of a downround or a poor share price at IPO if they fail to execute.

In walking though this daydream, I’m starting to think a founder’s willingness to take the money will boil down to a confidence matrix, something like this:

Huh! Well gee whiz, I think I can automate some predictive startup maintenance here (thanks again, Stanford!):

If you don’t have proof of market pull, you shouldn’t raise ultra large funding rounds.

Good founders solve problems in the market. Hopefully they also know what they’re doing, know what they don’t know, and appreciate the risks of high valuations before demonstrated market pull. Nest was a great example (or so the legend goes), and we coach our startups to think in this vein. Money can plug a lot of holes — especially in AI where top engineering talent is costing upwards of $1M/yr — but it can’t fabricate demand for a product or service startups are building.

If there’s demonstrated market pull, and you have the opportunity, take the money — with a mountain of cash, even those of us with low confidence in our capability to win can win.

But be warned! Fail fast and fail often is absolutely the way, but fail having spent $100M+ trying and you’ve probably done something horribly wrong. And the internet will make fun of you. With bigger buckets of cash flowing into the system, an increasing number of founders with the opportunity for big money will take an unnecessary risk.

Again, founders are paid at the exit, the burden of proof is on you. That proof is market pull. One of the great things about multiple fundraising rounds is that it’s a gate function for company growth; a check to filter out companies with innovations that have limited value.

Experienced VCs manning these gates can be complete assholes (I’ve been called worse), but they’re experienced assholes who’ve played witness to both the occasional rays of startup glory but mostly the eternal bloodbaths of startup war.

D-Fence! D-Fence!

$100M from one investor at early-stage isn’t defensibility.

One of the first questions we ask all founders we meet through Comet Labs is around defensibility — what makes your startup so great that others will have a hard time of being just as great? Some of the good answers are: access to unique data sets gathered by proprietary sensors, core patents developed over years and applied to a specific vertical, or demonstrated first mover advantage with the “data flywheel” making your ML-based solution better, and better, and better… Bradford Cross wrote an excellent piece about this last year.

What’s important here isn’t necessarily the validity of the defensibility, but the belief by founders and Comet Labs that such defensibility will hold over time.

It might not!

Things change, new sensors are built, another computer vision company focused on radiology comes along and poof! Defensibility is out the window. That’s one of the risks we’re willing to take, and why venture money (i.e. helping companies grow faster than without it) works.

An ocean of cash is not defensibility.

It’s hard to pick winners. Even with experienced founders, even with massive first mover advantage and demonstrated exponential traction, even with all the ingredients you could possibly hope for in a startup, failure is still the most probable outcome for any startup we invest in. We know this, and it’s why we fight like crazy to support our founders and their teams however and whenever we can beyond money.

When a startup takes in massive equity rounds of $100M+, that sure does feel an awful lot like the investors are picking winners. Some of these are easy, as in yes I think it’s safe to call ARM, NVIDIA, and WeWork as “winners” in the Vision Fund’s portfolio.

But the other companies are still too early in either their own cycle or the cycles of the greater market to definitively say “thou shalt be victorious.” And the Vision Fund jumping into early-stage is skewing perspective for founders.

This letter isn’t to me, SoftBank, or even other investors. It’s to the founders. Just keep building awesome things and try your best not to get sucked too deep into this strange universe where hypermassive venture rounds are a possibility. They’re not, nor are they necessary, nor are they valuable to you as a founder.

Lean and mean, baby. Lean and mean.

If this got you thinking and you’d like more, hit the follow buttons below and subscribe to the Comet Labs newsletter for a bi-weekly digest of our favorite AI and robotics stories!

Hypermassive Early-Stage Venture Rounds Increase Founder Risk was originally published in Comet Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.

Stefan Groschupf and the SalesHero team make difficult things look easy.

Exhibit A — Lots of people in San Francisco ride their bikes to work every day (myself included - watch yourself, MUNI drivers). Ben, the head of product at SalesHero, rides his bike to the office too. From Sausalito. Every day. No ferry.

Exhibit B — Stefan, SalesHero’s founder, was one of the original architects of Hadoop, and grew his previous startup, Datameer, from a concept into a data analytics powerhouse, raising over $100M in venture capital. Easy, no?

Whilst toiling on Datameer, Stefan stumbled on the serious need to improve his sales team’s workflow and efficiency for his company to survive. Selling new things is already tough. But for Datameer, selling visualizations is an art form. It was difficult on at the best of times, and a mean brick wall at others.

Enter innovation! What Stefan and his team at SalesHero set out to explore is a way to help teams capture the essence of software sales, giving teams new tools that let them focus on refining their craft.

Lots of SaaS companies have tried, and great progress has been made. But the problem still exists — it’s still really hard to sell.

The innovation at SalesHero is this: Getting humans to change habits completely is an ominous walk, so what if you could work that guidance into existing team habits and workflows with machine learning?

Enter SalesHero’s sales AI: Robin.

SalesHero offers the industry’s first personal augmented intelligence assistant that helps sales representatives simplify and improve their daily CRM workflow, automate tasks, enrich data and discover new prospects.

Robin takes on the tasks that constantly detract from actual selling by pulling relevant data from Salesforce, email, calendar and its proprietary database to automate sales tasks every day.

And Robin isn’t even the whole story.

Artificial intelligence will create unparalleled business opportunity in which AI augments human performance by allowing adopters to focus their attention beyond routine tasks.

It’s an enormous market opportunity that SalesHero is aiming for. Rescuing humans from tedious and repetitive work is the first step toward machine-augmented humans in the workplace. That’s the future that SalesHero is working to build, and we’re thrilled to be part of it.

SalesHero’s AI assistant rescues humans from tedious sales work. was originally published in Comet Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.

I like Y Combinator, a lot. In fact, I’m kind of jealous of Y Combinator. They have an amazing team, breathtaking network, $700M — $1B+ fund (depending on who you talk to), and the top founders from infinity and beyond banging down their doors to become part of that world.

I’ve also gotten to know some YC founders — excellent humans all. I’m even on record (well, at least a mental record) of talking my way into an epic fail with Justin Kan, a partner at YC. It’s a long story that he probably doesn’t remember, goes something like this:

• April 2014 — Andy is in Orange Telecom’s incubator, Orange Fab, and Justin comes to talk to the founders. I think there were four of us at the round table.
• Justin talks about his evolution from Justin.tv to something called Twitch.
• Surly Andy says something like “Im sorry Justin, but what the hell are you talking about? A streaming channel where people watch people play video games? Give me a break.”
• Justin laughed, and I think he said “you’re probably right!”
• August 2014 — Twitch sells to Amazon for $970M.

Sigh.

Back to YC through. There were also something like 140 companies — public and stealth — launching at YC’s 2018 Winter Demo Day.

One hundred and forty companies.

Thats a veritable smorgasbord for a rat like me! Getting a view at 100+ high quality companies in one go — when my job is to invest in high quality companies — is a hoot, an awesome challenge, and something that’s unique to the Valley. I love it.

But now that all of us at Comet had a chance to digest that kind of madness — along with a couple other demo days at Google, NVIDIA, and StartX (I’ll be back folks, this time from the other side of the table!) — we noticed that there were some common threads that didn’t sit particularly well with us. We want to share some reflections and talk through our experiences.

Statement 1: We’re 80% closed on our raise, and we need you to make a decision in 2–3 weeks.

This is something we heard many times, in the same vein of “I just graduated Code Academy, and I want $180k/yr, equity, and unlimited vacation to start.” The American dream — get what you can — but when everyone is trained to do it, the mantra loses its effect.

Getting the timing right for product/market fit is vital, and when a company nails that confluence of demonstrable prototype and peak market hype the results are powerful. As a founder, this was a great lesson for me — if you don’t naturally have that confluence, you have to do everything you can to create it.

Writeups in TechCrunch, pilots with Google, famous advisors and investors galore, founders talking shop at this event or that school…it’s all important and exciting. It’ll also probably get your company quite a few checks.

The intended strategy of cultivating artificial urgency through manufactured scarcity and peer pressure makes us think we’re about to miss something huge.

Maybe we are. But that’s just not how we invest.

From our perspective, a successful founder-investor relationship lasts for years, and we take that seriously. They don’t start under a time pressure.

Put another way, we don’t want to be sold. We want to believe. Belief takes time, especially if it’s something truly innovative and lacking the metrics to support a claim (which is a big part of the beautiful universe Comet blasts through on a daily basis).

Statement 2: We’re the (insert name of unicorn here) of (insert non-US location here).

These are actually interesting companies with the potential to move and grow extremely fast. Also, given Comet’s relationship with Legend Holdings in China, we realize that the ability to take demonstrated business models and apply them to a massive market like China is something we’re among the most well-equipped in the world to do.

But those aren’t the kinds of startups we invest in, and as a former founder it’s just not something I believe in either. And as soon as we made that decision inside Comet, we realized when assessing deals we’re always wondering if it’s based on innovation or entrepreneurship — two VERY different things.

An analogy we’re using in Comet to describe the difference between entrepreneurship and innovation is to imagine that technology is a highway we’re all traveling on that is always under construction. At any given time, there will be off ramps ahead of us, and there were off ramps behind us — depending on where and when we are, we can decide to take whatever off ramp we want to move onto whatever new roads lay beyond.

Innovation is building and upgrading exit ramps. It allows people and new activity to readily access a new road network. Sure, sometimes the ramp is unfinished and cars just launch off into oblivion, and sometimes there’s just a big wall that’ll just get crashed into (and again, and again, and again — just ask energy startup founders). But that’s all part of the process — innovation is just concerned with building or updating the exit ramp and enabling traffic to flow.

Entrepreneurs are the first to take the ramps to the bold new roads beyond (or smash into aforementioned wall).

We like this analogy because it a) clearly visualizes why timing is so damn important for entrepreneurs and b) helps nerds like me get over the fact that while we might be innovating really hard technology, it might not necessarily turn us into billionaires.

It also helps us define a scale to classify the kinds of startups we see. It looks something like this:

Comet tries to invest in “entrepreneurial innovators,” founding teams that have the vision of a new twist, technology, or trend and a strategy to capture the opportunity of the moment.

Innovation without entrepreneurship is academic. Entrepreneurship without innovation is sales. (Or piracy — arrg!) We don’t invest in either on its own. Again, we’ll probably miss out on good deals in hindsight. Selah.

Statement 3: We’re pre-revenue, raising $10+ M on $30M pre-money valuation.

That’s good work if you can get it, friends. But we believe this new flavor of fundraising is bad for the industry, for founders and investors alike. There are some that’ll disagree — well known names of the contrarian mindset. But we don’t have billions in the bank to fall back on once the hustle is gone, and neither do most founders.

I don’t need to describe why it’s bad for investors — its high risk without as big of a potential upside.

For founders, there’s only one word to be wary of: downround.

As a founder, it took me a couple years to learn this valuable lesson — the most important thing I can do is keep my company trending up. We should all fail fast and fail often, but it has to be in a way that’s constructive, educational, and moving us closer to our goals of taking over whatever piece of the world we’re after.

The problem with high valuation at an early stage is that it breeds skepticism. Skepticism begets the need for greater proof points to justify a high valuation. Raising lots of cash and avoiding dilution is the new Silicon Valley way, but you need to have the trajectory to match it and a solid understanding of what metrics you’ll need for larger investment to come into your company to fuel your growth down the road.

We believe there are two key trends that are driving this kind of fundraising:

1. A proliferation of new, well-funded VC’s focusing on seed and series A markets (guilty!)
2. Softbank’s Vision Fund.

There are a TON of new venture houses in SF — pure VCs, corporate funds, family funds, government funds, private equity playing in early stage, hybrid incubator-fund-strategic partners — all in a quest for innovation. Its competitive, and one of the few levers we can pull is offering better terms.

The Vision Fund is the more interesting thing, mostly because it’s still new and adds a different dimension to fundraising strategies.

It’s well documented that the amount companies are raising has been increasing over time. Fantastic, makes for a great party! But with one fund looking to invest $100 million at a minimum in around 500 companies in a short time just brings to mind those massive concentrations of gravity and contractions of the surrounding universe. Black holes… and more thoughts on that here.

Suffice to say, the Vision Fund establishes a third exit opportunity for founders and seed investors: IPO, acquisition, $100M in cash. As demand for $100M deal flow increases, the bar to achieve this kind of investment decreases — more sizzle less steak, a shift toward the entrepreneur who sells dreams before building them.

It’s a brave new world for sure, and adds an exciting new dimension to this wild startup universe. But it’s important to remember that something like the Vision Fund is a startup in itself, the biggest startup ever funded in the history of humankind — bigger than NASA, bigger than the grid. Pretty radical, definitely bold, and absolutely with a good probability of failure. Welcome to the startup club, Vision Fund!

But for founders, why can a high valuation too early be a bad thing?

It all comes down to probability. If you raise at a high initial valuation, what is the probability that you’ll be able to secure additional future funding at a yet higher valuation? The first question I ask when I see high valuations is “what milestones are the next round of investors expecting you to hit?” In most cases, the answer here is vague — and that’s a challenge.

Yes, one venture model is to try to find that one company out of a hundred that’ll be huge — forget the other 99, we have to go BIG. We’re all in it to be that one person, that one company who makes it huge. But the reality is that we’re not all that singular beast, and we’re not all riding that rocket ship to unicorn land . . . yet.

Startups and VCs only work when both see everything as a partnership, a symbiotic relationship where each is enhanced by the other. For us, that means that we see investments as more than a bet on the future. Each one is an investment — in money, time, and help — that deserves our attention and energy. It’s a value that drives our investments into Entrepreneurial Innovators, and we’re proud to stand by it.

If this got you thinking and you’d like more, hit the follow buttons below and subscribe to the Comet Labs newsletter for a bi-weekly digest of our favorite AI and robotics stories!

Three things I hear from founders that make me nervous. was originally published in Comet Labs on Medium, where people are continuing the conversation by highlighting and responding to this story.