Two years into the AI wave, we thought it will be price talking to an early-stage investor in the industry to get an inside look at what it’s like to speculate in generative AI startups from day one.
For this purpose, we caught up Costanoy’s ventures founder Greg Sands and general partner John Cowgill. The Palo Alto, California-based company is investing in pre-seed, seed and Series A funds and closed its fifth $275 million fund (the largest) and its third $119 million opportunity fund to speculate in its winners, each in September.
Since its first $100 million fund raised in 2012, the company has grown its practice in enterprise cloud services powered by data and analytics.
Costanoa currently focuses on applied artificial intelligence and AI infrastructure and B2B fintech. With generative AI, “we can solve problems that were unsolvable two years ago,” Sands said.
Costanoa is also increasing its cybersecurity efforts – as threat incidents increase with artificial intelligence and the potential to enhance security operations – and national security. The company incubated Vannevar Laboratories in 2019, before defense technologies and artificial intelligence were as popular as they are today.
New possibilities that you just have never dreamed of before
Advances in artificial intelligence in recent years mean that, due to natural language processing, written language will be easily digested, consumed, organized and even reasoned about. Computer vision makes it possible to judge and understand images. Combine these technologies and you begin to get some very interesting and complex results, Sands said.
Going vertically
“We have literally solved most of the problems that this generation of technology could solve,” Sands said of the SaaS revolution that has taken place over the past 20 years.
The company believes that generative AI has now opened up vertical opportunities that will have been too small in the past.
“He’s always had a knock on the SaaS industry: It’s a smaller market to reach because you’re limiting yourself to one industry rather than building something that can be applied to every industry,” Cowgill said.
The company looks closely at each industry to grasp how big it may change into as artificial intelligence develops.
In this spirit, Costanoa supported this decision Aquabytewhich provides computer vision for fish farming. She also invested in Traffic light artificial intelligence for the help of pilots in aviation and ForceMetricsa service that gives first responders with situational awareness through data consolidation and presentation. He also made a secret investment in a company that issued building permits. While this may increasingly seem narrow, there is a strong opportunity to have a comprehensive workflow in this space.
Bullish on native AI
The big query for investors is: does this vertical opportunity profit existing API-enabled mid-market portfolio corporations? Or is this paying homage to the cloud and mobile era, where rebuilding business models was a challenge for incumbents?
“I’m more optimistic about native AI application layer companies than I was two years ago,” said Cowgill, who has seen firsthand how difficult it is to integrate AI and what the advantages are.
The company has several vertical later-stage SaaS corporations working on AI integration. “The pace of AI integration will be slower,” Cowgill said. “They may actually provide different services and different products, different business models.”
“The real value of AI comes when you can have an end-to-end workflow,” he said. “This is the idea of an agent that steps away from using AI, points to something and says, ‘search this with AI, summarize with AI’ — to have a task to do with AI. Getting agents to work is extremely difficult.”
Nuances in picks and shovels
Costanoa has made several bets in AI infrastructure. It invested in Dolphincreator of junior AI data scientist who helps teams. This was also supported Rerun.ioopen source visualization stack for multimodal data including audio, image and video, and OpenPipewhich is used to tune models.
Still, Sands noted that model management infrastructure has generally not developed as much as expected 18 to 24 months ago, in part as a result of the undeniable fact that AI corporations do not manage models themselves but purchase from hyperscalers .
“Things that matter”
But some things have not modified.
In a 2018 blog post, “What is $%&* applied artificial intelligence?Sands noted that AI does not replace the need for “excellent product management” and that “data is as important as the algorithm.”
Sands said that throughout the company’s 12 years of existence, he has focused on age-old principles: “find extraordinary people and give them the time, energy and opportunities to give them the best chance of success,” he said. “Choose things that matter – both technologies that matter and problems that matter.”