Agile development, rapid learning, granular instrumentation, a large-scale compute and data infrastructure, cutting-edge machine learning — these are the ingredients that give innovative companies their edge.

But to date, this mindset and approach have been largely confined to the world of software and Internet-based services. Look outside of that space and you’ll find a wide range of industries — some hundreds of billions of dollars in value — that are slow-moving and starved of rapid iteration and learning. Think aerospace, transportation, manufacturing, logistics, material sciences, life sciences, and agriculture.

These industries are ripe with opportunities for transformative startups, and are seeing a new wave of entrepreneurs capitalize on them by unlocking the agility and speed of learning we take for granted in the software world.

We’re still in the early days of this movement, and the specifics will vary by industry, but we’re beginning to see the contours of a “playbook for transforming big industry” emerge:

1. Take an end-to-end approach to a product/service. By controlling the stack, startups win the ability to aggressively experiment with production processes and tighten feedback loops.

2. Dramatically reduce cycle time for learn-build-test loops. By employing agile techniques and automation, startups are shrinking their iteration loops and accelerating their learning curve.

3. Carefully instrument workflows to collect data. By capturing the right data (or new data altogether via cheap sensors), startups are fueling much stronger feedback loops.

4. Apply machine learning to close the loop. By capturing more data more frequently and leveraging end-to-end control to directly feed learnings back into their systems, startups are improving their products/services and their cycle times to create powerful virtuous cycles.

Let’s look at a few concrete examples of this playing out in aerospace, life sciences, and agriculture.


For decades, satellite imagery has required huge, complex satellites that orbit far above the Earth. This approach has required heavy investments in time and capital to develop and launch these satellites, leading to cycle times on the order of years, if not decades, between new satellite hardware, imaging capabilities, etc.

A new cohort of next-generation satellite startups has been turning this model on its head. Instead of spending more time building a limited number of expensive satellites that orbit for a long time, they iteratively build many low-cost satellites that orbit for a shorter amount of time. This enables them to offer global coverage on a daily basis, iterate their offerings more rapidly, and spread the risk of launch failures. Here’s how we think about all this in the context of the playbook described above:

1. These startups take an end-to-end approach to delivering imagery and intelligence services, controlling everything from the satellite hardware and software to the data feed and apps/products that apply the data to benefit end customers.

2. By controlling the design, development, and deployment process in house, they dramatically reduce cycle times involved in the complex coordination across hardware, software, and services.

3. Again, by controlling the entire process, they can also instrument everything from the satellite hardware and software to the data feeds and apps, collecting proprietary data to power feedback loops.

4. Armed with dozens of satellites producing global coverage and daily imagery, they can exploit machine learning to optimize their overall core value proposition and operational efficiencies.

Spire and Planet are two great examples of companies using this exact approach to offer new services ranging from agriculture and defense monitoring to maritime and weather data. Rocket Lab follows a similar playbook towards a different but equally compelling end: affordable, high-frequency rocket launches. As these companies scale, their impact could extend well beyond their immediate industries and pave the way for space-enabled consumer services such as ubiquitous Internet access, “dead zone”-free communications, or even virtual tourism.

In three years since Planet launched its first satellite, it has designed 13 different builds of the Dove satellite. Seen here is an evolution in the design of the Dove, with seven of the most significant build iterations.

Above: In three years since Planet launched its first satellite, it has designed 13 different builds of the Dove satellite. Seen here is an evolution in the design of the Dove, with seven of the most significant build iterations.

Image Credit: Planet

Life sciences

The mental image that likely comes to mind when you hear “biology research” is strikingly close to reality: PhDs running around labs, pipetting precise volumes of liquids into test tubes, moving microwell plates from one piece of equipment to another, and manually entering data into lab notebooks. Wet lab experiments, in particular, rely heavily on human processes and outdated equipment. As a result, they are slow, expensive, and very difficult to reproduce — making a fast, stable feedback loop challenging, if not impossible.

One area where startups are upending the status quo is in engineering microbes to increase the supply of biological products. Companies like Gingko Bioworks and Zymergen are transforming the traditional process by automating high-throughput experiments to construct and evaluate thousands of strains in parallel to find the best one and then learn from the vast amounts of data generated to optimize future experiments. Here’s how they fit the playbook:

1. They take a (mostly) end-to-end approach to offer engineered microbes as a service to their partners, controlling the labor, equipment, and experimental processes.

2. They fully automate the lab environment and optimize for high-throughput experiments to dramatically increase the speed, quantity of data, and learnings relative to traditional experiments.

3. They also instrument and collect data from every step of the experimental process, which provides the benefit of consistent reproducibility.

4. They apply machine learning to the data collected over time to optimize and select strains for future experiments, helping drive both improved production and more efficient experiments.

In the near-term, we will get lower cost, higher quality, and more environmentally friendly compounds for use in products ranging from textiles to food. Bolt Threads has developed a unique method to naturally produce fabrics for consumer apparel. Startups such as Clara FoodsGeltor, and Perfect Day are at the cutting edge of engineering microbes to create animal-free food products – respectively: egg whites, gelatin, and milk – resulting in many new, healthier dietary alternatives. In the long-term, applying state-of-the-art software and automation to engineering biology will lay the groundwork for dramatic progress throughout the life sciences.


Agriculture is a particularly interesting opportunity given the industry’s reliance on external and uncontrollable forces like weather – a pretty fundamental constraint. This combined with relatively low and fragmented adoption of automation and data-driven systems severely limits learning velocity and agility.

“Controlled environment agriculture” is emerging as an economically feasible solution to these constraints, thanks to advances in LED lighting efficiency, sensor and compute cost/performance, and data science. A new wave of indoor farming companies such as Freight Farms and Spread have been taking this exact approach. And once again, we see the playbook at work:

1. They take an end-to-end approach — controlling everything from growing platforms to inputs to computing infrastructure to climate systems — to produce crops they can sell.

2. The controlled environment eliminates seasonality, leading to far more crop cycles per year and enabling more parallel experiments with different growing processes.

3. They also instrument the entire growing process and collect millions of data points on each plant in every crop cycle to understand how to optimize the growing process.

4. They apply machine learning to data collected over far more frequent and better instrumented crop cycles to improve both growing outcomes and environments.

As these efforts scale and the benefits of agility are fully realized, agriculture will begin to look more like next-generation manufacturing — automated, precise, and optimized — with profound impact on humanity and the planet. The industry will provide better food quality, affordability, and access for consumers, while also using our planet’s precious natural resources far more efficiently.

This is the right time to jump in

There’s an incredible period of transformation ahead for industries with historically limited learning velocity and agility. We now have both the will and the means to fundamentally rethink the systems we rely on for transportation, materials, healthcare, food, and so on. As founders and investors take advantage of the present opportunities, the future impact will be tremendous. The learning velocity will continue to accelerate, resulting in extremely agile technology companies delivering better, faster, and more cost-effective offerings to climb the ranks of all industries. Now is the time to embrace a new agile playbook or risk getting left behind.

[Disclosure: Our company, Innovation Endeavors is an investor in three of the companies mentioned in this post: Planet, Zymergen, and Bolt Threads.]

Dror Berman is managing founding partner at early stage VC firm Innovation Endeavors, and Samantha Wai is a member of the investment team at the firm.

Get more stories like this on TwitterFacebook