If you buy something using links in our stories, we may earn a commission. This helps support our journalism. Learn more. Please also consider subscribing to WIRED
A few years ago I had a jarring discovery. I’d spent more than 15 years building and launching technology products, but all of a sudden the skills and instincts I’d honed seemed irrelevant. In fact, if I stuck with them, they’d likely lead my company to ruin. That’s the realization I had with my company Lytro, which is building a new kind of camera and software for creating movies, TV shows, and virtual reality experiences. With Lytro I’d embarked on a new kind of challenge — that of a “hard tech startup.” These are companies where, at the outset, you don’t know whether the core tech needed for your invention will work at all. The forces of physics, biology, and Moore’s Law can bedevil you in unexpected ways. Building a winning product is already challenging enough; with hard tech startups, the obstacles are even greater.
Our goal at Lytro is to unlock new creative freedom across VR and traditional film and television. With Lytro Cinema, this means letting directors and cinematographers control more aspects of their shots — such as focus, exposure and frame rate— computationally in post-production. (Historically, they had to commit to these choices at the time of the shot.) We had to develop the world’s highest resolution, highest frame rate image sensor; invent a new kind of optics; solve enormous data and storage challenges; and create a whole new class of imaging algorithms. When we started out, we had no idea if any of that was possible.
Companies facing that kind of challenge have to throw out a lot of conventional startup wisdom. Over the last decade, now classic books such as Eric Ries’ The Lean Startup have become the default sets of principles for building successful products and companies. With a rapidly growing global Internet population and a dramatic reduction in infrastructure costs (such as through Amazon Web Services), an entrepreneur could rapidly and inexpensively iterate an idea to achieve product/market fit. The explosion of smartphoness and the emergence of the social web amplified distribution, allowing young companies to much more easily reach their target customers. But those rules of thumb don’t apply to hard tech startups.
Through this experience, I’ve come to adopt a whole new mindset. I’m hugely excited about this evolution in startup thinking and believe we are on the verge of turning ideas that were long the subject of science fiction into reality. Some of today’s most vibrant areas of entrepreneurship are in hard tech areas, including autonomous vehicles, space, artificial intelligence, virtual reality, and genomics. The output of these companies may be even more dramatic and profound than the build-out of the internet has been.
Below are the lessons we’ve learned so far about how to build a hard tech startup. Although the examples are specific to Lytro, I believe they apply more generally to companies trying to build anything radically new leveraging cutting edge technology.
A popular approach to product development is to build what’s known as the Minimal Viable Product (MVP). The goal is to develop just enough functionality to see if a product resonates with an early group of customers. This process is usually fast and capital-efficient and allows you to iterate your way to success.
That doesn’t work with a hard tech product. Here, you’re often, if not always, tackling a problem that no one on the planet has ever successfully solved. So you have to figure out the smallest, fastest, most capital efficient way to confirm that what you’re proposing is possible.
A perfect example arose for us in early 2015, when we started trying to build the light field image sensor at the heart of our system. A light field camera captures not only the intensity of light, as a traditional camera does, but also other features of the light in a given scene, such as the direction it’s traveling.
In order to achieve our product goals, we needed a 755 megapixel (MP) sensor capable of recording at 300 frames per second. That’s roughly 90 times the current industry standard of about 8.3 MP. Building it ourselves would have taken more than three years and cost us over $10 million — if we could even do it. So first we needed to know if the core idea was even technically feasible.
Here’s a visualization of the Light Field and how we capture it.Rather than trying to build this sensor of our dreams from the ground up, our team, led by the tenacious Jon Karafin and Brendan Bevensee, decided to combine many off-the-shelf image sensors, in essence meshing them together into a giant monolithic sensor to reach our 755 MP goal.
Step one was to determine if we could make two of the many sensors work together reasonably well. This alone took quite a bit of supply chain and calibration work. Here’s the first test image the team produced:
Lytro VP of Business Development Jim Migdal reluctantly posing for the first Lytro Cinema test image.The results of this MVTV gave us the confidence we needed to proceed to the next step which was to build a small-scale, low-resolution light field video capture system. While we kept the resolution the same as our MVTV effort, we added all of the additional software and hardware that turned it into a light field system. Our code name for the overall project was Trillion, so we named this prototype Mini-T.
Mini-Tin the lab and in the wild for a test shoot.Using Mini-T, we developed a set of demos and proofs of concept that gave us the confidence to proceed with a full system build.
Our very first, low-res light field video showing off some of the computational capabilities of Lytro Cinema.We then spent another eight months building out the first version of our product, which we launched at the National Association of Broadcasters (NAB) show in April 2016, a mere 12 months after kicking off the project.
The room was packed beyond capacity at the Lytro Cinema announcement at NAB 2016.
A visual effects breakdown from the short film “Life,” shot on Lytro Cinema.
One of the biggest challenges in doing a hard tech startup that fuses proprietary hardware and software is the fact that the critical components likely don’t exist in the form you need — if they exist at all.
When I started at Lytro, I naively thought that solving this problem was as simple as going to Shenzhen, China, or some other manufacturing center, and talking to the leading suppliers and contract manufacturers to either find something off the shelf or convince a supplier to tweak an existing part. After all, why wouldn’t these suppliers want to work with a Silicon Valley company that was building something cool and groundbreaking?
Shenzhen, China, manufacturing capital of the technology world. (Daniel Berehulak / Getty Images)It turned out I was dead wrong. Figuring out which suppliers offer something close to what you’re looking for is in itself a dark art. Large tech companies often use their market power to ensure that that no one else can access particularly important parts of their supply chain. This strategy has been a hallmark of the smartphones industry, where competition over displays, lenses and optics, housings, and other components can differentiate a product from its many competitors.
Going through this process the first time was terrifying. We had to find just the right combination of image sensors, camera electronics and optics. On many occasions I thought we would never find it, and our path would end.
Testing one of the many image sensor and camera electronics combinations we tried before finding one that worked.Over time, I came to see this challenge as a fun part of the invention process. It became both a treasure hunt for the right suppliers and a puzzle, as we learned to evolve our design to meet the parts we managed to find. Solving this puzzle the first time often unlocks a whole set of new ideas for improving the design, and can become a source of competitive advantage for a company. In our case, the suppliers that best suited our needs turned out to be in the United States and Europe rather than Asia.
So, you’ve built a successful prototype of your game-changing product. All the key features work as you’d hoped or even better. Your first beta customers are blown away. The belief that you can change an entire industry or the way people live is strong. The only thing standing in your way is taking your prototype and turning it into a robust product, ready for the intense scrutiny of your customers.
Predicting the amount of time, effort, and capital you’ll need to go from prototype to product in any technology product is tough. But in my experience, doing so in a hard tech product cycle is particularly difficult. A good rule of thumb is to take your most conservative estimate of the required time and cost, and then double it. The reason is that — again — you’re typically trying to build something that, at both the hardware and software level, has never been done before and falls squarely into the category of unknown unknowns.
In the case of Lytro Cinema, we found that just about every part of the system, including mechanicals, electronics, and optics, needed to be significantly reworked at least once if not multiple times in order to meet the requirements of our future customers. (Our target customers are big-budget films and televisions shows. Not surprisingly, these customers are some of the most technically and creatively demanding in the world.) This, of course, is on top of the massive amount of software development and calibration we needed to tackle to deliver on our target feature set and image quality goals.
Many first-time hard tech entrepreneurs dramatically underestimate the time and cost of this phase. Based on my conversations with a number of founders who have launched their projects on Kickstarter to huge excitement and pre-orders, this part of the journey — from prototype to full product — is where so many of these projects slip months or even years beyond their original projections.
Building a hard tech startup lets you work on an intellectually interesting set of problems with the potential for huge societal impact. The competition is often less fierce than what an entrepreneur typically sees in more mainstream product categories. But it is a patient person’s game and goes against the Silicon Valley mantra of hypergrowth. When building a highly disruptive product or company that at its core relies on radical new technology, it’s crucial to select your initial customers with care.
For us that means focusing intensely on the success and happiness of our first few customers as the foundation of our business. Starting with 20 or more customers, with some portion of them succeeding while others failed, would have been a much harder and riskier path. Why? Because those first happy, successful customers will be our best allies in signing the next customer and the one after. Giving ourselves the time and space to learn from our first customers, account for the unknown unknowns, and implement the inevitable set of required and desired enhancements they will need is much easier with a handful of customers than with many.
Every era of technological innovation builds off the one that preceded it, and today we have an increasing number of entrepreneurs tackling truly ambitious, world-changing problems. The combination of low-cost cloud computing, a renaissance in machine learning, and rapid advances in genomics have opened up new classes of problems for computers to try to solve. The accessibility of the global supply chain, driven in large part by the smartphones industry, has made it faster and cheaper than ever to manufacture physical products. Hard tech startups dive into these new opportunities, often fusing hardware and software or software and biology at a level of sophistication far beyond what any entrepreneur would have dared tackle even 5 to 10 years ago.
To succeed in these areas, we need to update our collective wisdom of how to build successful companies. This is our theory of victory — a first draft of that new playbook. We’re still very early in our journey with Lytro Cinema, and I have no doubt we’ll learn more hard lessons along the way.