Original Thoughtful Circle Thoughtful Circle
Today, the veteran fund NEA published a new article: "Rewriting the Startup Playbook for the Age of AI", which focuses on the experience of the founders of four generative AI companies invested by NEA: Perplexity, Rewind, Forethought and Crowdbotics. These three founders summarized four new trends in startups in the AI era, and a lot of content should be of great help to AI entrepreneurs, and the author quickly translated this article with the help of GPT-4, hoping to bring you the latest thinking and inspiration.
In addition, the previously created deep thinking circle information sharing group has been running for almost half a year, which will share the latest useful and interesting information in real time, especially AIGC and overseas operation growth, the information density and freshness in the group is very high, the communication atmosphere is also very good, everyone discusses and learns in the group. At present, 1 group, 2 group, 3 group, 4 group, 5 group, 6 group and 7 group are full, I have just created 8 new groups, welcome interested friends can reply to the keyword [information] in the background, scan the code to join the group.
Thoughtful Circle's 44 original articles on learning, work, life, investment, entrepreneurship and life
Official account
Lead
With the explosive growth of generative AI breaking the rules of company building, the NEA-backed founders identified four key emerging trends.
Interest in generative AI reached its peak last fall, when Perplexity AI co-founders Aravind Srinivas and Denis Yarats knew they had to move fast. They don't have weeks or months to recruit, so they abandon recruiters and schedule interviews and cultural fit discussions. Instead, they take a more direct approach: They offer a two-week paid trial to their preferred candidates.
The plan worked, and Perplexity hired co-founder Johnny Ho, who joined the company just days later as chief strategy officer. With the exception of the fourth co-founder, Andrew Konwinski, all of the company's employees were hired this way. In addition to saving time, the process weeds out those who would otherwise be unlikely to join and inspires enthusiasm for Perplexity's mission better than any presentation or incentive package. "Now I'm hearing that other companies are also using this trial-based approach to hiring," Srinivas says. ”
Since the launch of the viral chatbot ChatGPT last November, venture-backed startups like Perplexity have had to abandon the traditional model of some company building and embrace new ways of thinking. NEA Partner Ann Bordetsky said: "From fundraising to product development to customer service, many time-honored practices will become obsolete in just a few years. We're in primordial chaos right now, and things are a bit confusing and experimental. But you can see that people are building new approaches. ”
This isn't the first time a breakthrough technology has sparked a profound change in how great companies are built. Like inventing the personal computer, founders like Steve Jobs and Bill Gates created new rules for manufacturing, marketing, and selling complex digital technologies. So did Amazon, Yahoo and countless internet startups in the '90s, Facebook and other social networking companies after the rise of smartphones, and Salesforce, Box, and other SaaS providers that turned to the cloud.
As one of Silicon Valley's most established venture capital firms, NEA has been at the forefront of these revolutionary technological shifts. If there's one thing the company has learned over the past four decades, it's that chaotic times like the current one tend to spawn some of the world's most successful and influential companies.
"These technologies all enable previously impossible business model innovation," said Scott Sandell, NEA's chairman and CEO. For example, the Internet has made open source feasible and made it possible to deliver software as a continuous improvement service rather than a licensed product – free if chosen. "It changed the way you develop software, distribute software, and get paid for software," Sandy said. "I believe AI will have such an impact as well."
NEA partner Aaron Jacobson said that if anything, the impact of AI would be even greater. While previous upheavals have involved how and where technology is used, "AI is actually changing who does the work," he said. "It's unprecedented, so this disruption will be faster, more violent and bigger than ever before, because there's more to fight for."
The generative AI boom is less than a year old, but a lot has already happened, and we can already see how company building will change in the coming months, years, and possibly even decades. To better understand this historic shift, NEA asked four founders of AI startups in its portfolio to share their new thinking. From these discussions, we have identified four new trends in company building in the AI revolution.
01.
Trend #1: Flexibility matters
Even compared to past booms, the generative AI market is moving at an astounding pace, with some incredible new ways to use large language models (LLMs) being released every week. Tech giants and leaders such as OpenAI, owners of Google, Microsoft and ChatGPT, are pouring tens of billions of dollars into not only creating these models, but also developing APIs and other tools to help innovators commercialize their products, often preferring to embrace open source rather than pursue a closed-garden strategy in search of lock-in and higher margins.
In addition, generative AI allows companies — not just AI startups, but all types of companies — to move faster. When Srinivas left OpenAI to start Perplexity AI, the first-time founder used the company's own "answer engine" technology to learn things that would take hours of Google searches and countless lunches and coffee conversations with experts. "We don't know a lot of the basics that founders need to know, like how to do corporate taxation. We don't know you have to issue a Form 1099 to the contractor," he said.
As a result, companies are executing at breakneck speed. The team released four chatbot versions of OpenAI-based GPT-3.5 LLM Perplexity within a few months, attracting more than a million views per day. Then, when OpenAI released GPT-4 in mid-March, co-founder Yarats quickly convened the entire company for a long hackathon. Two weeks later, they launched Perplexity Copilot, an "interactive search partner" that could ask users further clarification questions, conduct multiple searches at the same time, and provide more accurate results.
Srinivas noted the success of fast-iterating AI startups like OpenAI and Midjourney, "those who execute fast are getting paid, and speed is a constant here." ”
And speed requires a different mindset. At a recent all-hands meeting, when an employee asked if he could set measurable quarterly goals, Srinivas said he was happy to consider — as long as employees were prepared to change their goals every few weeks.
This flexibility also requires a more flexible approach to technology and innovation, and companies need to maintain strict neutrality on the technology they use to develop their products, even with the products and technologies they create, rather than betting on specific suppliers or proprietary products. When Deon Nicholas saw advances in the underlying model in 2017, he founded Forethought, an AI-based customer service system. Due to breakthroughs such as OpenAI's GPT-3.5 and ChatGPT in 2022, the company has consciously abandoned most of its self-developed technology stack.
Nicholas said: "We went through five stages of grief because we had several years of technical barriers. But eventually, we realized that we could continue to be two or three years ahead of the competition by applying technology like GPT-4. AI will fundamentally change our market, so it's important to lead that change. ”
Adaptability is necessary when Dan Siroker, founder of Rewind AI, offers a service that gives people "perfect memories" through searchable records of all digital interactions. "The world is changing at an ever-increasing rate, so the ability to be faster than everyone else is more important than it used to be," he said. That's why we release 11 versions of our products every day. I would even say that your ability to react and adapt is more important than your ability to predict the future. This may have been the trait of a great founder 10 years ago, but I don't think it's as important as being able to pay attention and listen every day and make better decisions. ”
02.
Trend #2: Reimagine productivity
ChatGPT's consumer adoption rate makes TikTok seem slow, with millions of people around the world raving about how it can boost their productivity. Leading startups have begun to implement changes that allow them to do more with less.
This productivity boost creates the basis for faster growth. First, it makes the company more profitable. An insurtech startup jumped from 40% to more than 50% after some lightweight training on its large language model.
Generative AI helps startups achieve their rapid growth while maintaining the advantages of being small, such as flexibility and team spirit. For decades, research has shown that small teams are more efficient and productive when it comes to developing software. That's why Perplexity received $26 million in Series A funding, though it could have raised more money: to prevent the company from growing its headcount faster than absolutely necessary. "A special kind of magic happens when there are the right people doing the right things," Srinivas said.
This magic is also important for recruitment. "There's no better way to market yourself to employees than to show your rapid progress," he said. "You can say anything you want about the roadmap and vision, but why should anyone trust me, the founder of a first-time startup? Because the team releases a great piece of software every few months. ”
Ultimately, the company's product strategy also needs to reflect this productivity revolution. "Seats"—licensing software according to the number of people allowed to use it—as a way to sell business-to-business software. "If the amount of work done by each employee increases by a factor of 10, you'll want to focus on the output, or any other unit of value that the customer cares about," Siroker says. ”
So, how lean is enough? There are no clear guidelines yet, and too many things are still changing. But headcount will be drastically reduced, said Siroker, who founded and sold digital experience platform maker Optimizely before founding Rewind. Companies are already changing their hiring targets to select more generalists, not because they have deep expertise in a particular programming language or task, he said, and LLMs are learning those tasks quickly. When everyone in the company is using generative AI, "it's like they're all wearing Iron Man suits." ”
"We're building an amazing company with 15 people," Siroker said. In the past we may have needed hundreds of people. It's a different way of thinking. ”
The result will be a new breed of very valuable, very small companies. NEA's Jacobson said: "We're going to see companies that generate hundreds of millions of dollars in revenue with just 25 or 50 people. It would be crazy. ”
NEA partner Vanessa Larco said this hyper-high productivity could lead to new problems, including a massive escalation in the battle for top AI talent. OpenAI has reportedly paid more than $1 million in salaries. "Generative AI may turn a 10x engineer into a 100x engineer, but it won't make a mediocre engineer better. It's like you give a calculator to someone who isn't very good at math," she said. "They don't know how to make the most of it."
03.
Trend #3: Build data barriers
Network effects, the wonderful phenomenon in which the value of a product or service to a user grows with each new customer. Facebook is more valuable to people with 3 billion other users on the web than to people with 30 or 3 million users, and companies with huge market capitalizations almost always have network effects.
Success in the age of generative AI will depend less on who has the most customers in the first place and more on who can figure out how to get more of the data they need to build better products than the competition. "It's all about data," Larco said. "You can be the first move, but if you don't take proprietary data sets and gobble them up in bulk, it won't help to start early."
Therefore, founders in the era of generative AI need to see data as a more important strategic priority. After all, other traditional sources of competitive advantage may not be viable. Given the innovation boom around generative AI, no startup is likely to maintain a big technological advantage for long, especially when relying on a popular foundational model like GPT-4 that is open to everyone. Forethought's Nicholas points out that for AI-based products, the best branding and marketing is the correctness and reliability of that intelligence. "It's almost impossible to erect barriers in traditional ways, such as technology or branding, and the only real way to do that is to use proprietary data."
For consumer companies, success depends largely on a truly disruptive user experience. "What we're looking for is virality," Bordetsky said. Perplexity AI's daily traffic, for example, is partly due to its "answer engine" that not only provides ChatGPT-style query answers, but also links to information sources.
And, these user experiences need to be designed with data in mind. "Companies need to be particularly good at aggregation, user growth, and engagement to continuously improve their AI-driven products, because that's what really drives the business," Bordetsky said.
Since founding Forethought in 2017 to create better customer service chatbots, killer user experience has been a focus for Nicholas. The company invested in pipeline software to easily integrate data from existing systems like Confluence, Salesforce, and Zendesk to inform its chatbots. And in March, it launched an OpenAI-based service called SupportGPT.
As a result, the company now has more than 50 integrations, which has helped it sign up more than 100 customers, including Marriott and Instacart. Nicholas says the company is building a reinforcement learning system to capture even seemingly trivial interactions, such as whether a help desk agent actually used a recommendation from the Forethought system and whether the end customer was satisfied with that answer.
"Ultimately, it's about creating an intelligent system," he said. "To build a customer service operation today, companies hire agents and equip them with customer data from traditional systems of record." In 10 years, companies will first use AI to build a system on top of that proprietary data, so a help desk agent can simply ask, 'What was our interaction with Deon a few weeks ago?' ’”
04.
Trend #4: Think bigger
No technology in history has emerged as quickly as generative AI — not search, not smartphones, not social media. Unsurprisingly, the level of competition is rising just as fast.
As of May 2023, Dealroom counted more than 250 generative AI startups, and tech giants are eyeing the technology as much as their future depends on it, which may indeed be the case. And, because they already have a lot of valuable data, these incumbents have an advantage over their predecessors in fending off the next generation of disruptors. Given this turbulent environment, "it's not a time for incremental thinking," Larco said. "You need to do something so different and obvious that people look back and say, 'I can't believe we used to do that.'" ”
This means moving away from piecemeal product improvements and instead focusing on creating entirely new categories. For Dan Siroker, the founder of Rewind, his goal is to bring people "perfect memories." Once a person agrees to give them access to digital activity — the websites they've visited, the texts they've sent, what was said on a Zoom call — Rewind's app can retrieve any interaction, even if the person only remembers a few key words that were typed or spoken. (For now it only works on the latest Macs, as they rely on Apple's M1 and M2 chips.) ）
The plan calls for inventing a compression technique that compresses data more than 3,000 times, making it possible to store years of data on a user's device. They must also figure out safeguards that deal with obvious privacy concerns (e.g., all data remains on the device and is never stored in the cloud).
However, strong challenges are part of the appeal. "I see this more like a problem I want to spend the rest of my life solving than a business plan," said Siroker, who likened Rewind's memory-keeping abilities to a "superpower," similar to a hearing aid that restored hearing when he became deaf in his 20s. "It's more about selfishness than strategy."
Siroker carefully found investors he was sure would support his mission. This led him to a group led by the NEA who praised our track record of investing in companies in all the inevitable ups and downs — sometimes even buying rather than selling at IPOs. "The people at NEA understand the odds of successfully completing this task, and they are willing to participate," Sirker said. "They are interested in building a company with long-term value. This is one of the reasons why they are one of the few companies that have been successful in every generation of technology. "
Siroker agrees that executing a bold business plan is easier — not harder. "The bolder the idea, the more excited employees and potential investors are," he said. "It also helps me stay excited about what we're doing."
Perplexity's Srinivas agrees that true Great Commissions can inspire great people, but stresses that this needs to be underpinned by real achievement. "I'm realistic that we can't compete with Google on pay. No one can," Srinivas said. "But right now there are a lot of very talented technologists who are extremely bored in places like Google, and they're looking for opportunities to make their mark on the world."
Crowdbotics is another NEA portfolio company that uses generative AI to pursue lofty goals. Simply put, the company intends to use AI to reinvent the entire software development process. While a large number of low-code and no-code competitors are creating tools so that non-technical people can create software for simple applications, founder Anand Kulkarni believes that the foundational models that began to be discussed in academic journals in 2016 will one day allow application development teams to do more work with natural language commands.
"Any software engineer will tell you that writing code is the easy part," Kulkarni said. What's more difficult is figuring out what the software is supposed to do, and how to express that in terms that computers can understand — not to mention despised tasks like writing documentation and making sure the generated code is safe. In addition, Crowdbotics has developed processes that allow companies to save all software produced by their developers to a catalog of reusable components.
Ultimately, his vision is to "reduce the marginal cost of developing software to zero" by having companies produce production-ready code in hours or days, rather than the months or quarters it takes now, Kulkarni said. In a decade, developers will be able to describe the software they're trying to create, and the company's "code manipulation" system will stitch together existing components, help developers write the rest, and build the necessary security, privacy, and bias fences.
This process will never be fully automated, as humans will have to review and rule out corner case issues. Even so, this approach has unleashed the potential power of innovation for Crowdbotics' customers. For decades, most innovative software-based ideas have been rejected, often because of concerns about engineering costs. "There's so much friction in pushing ideas to the starting line that most ideas are never considered," Kulkarni said. "We're changing these economics by making it efficient and easy to operate."
Not so long ago, Kulkarni didn't declare this mission too strongly. The company played down its use of generative AI for fear that potential customers would scoff at the idea that AI could handle something as complex as building enterprise software. "We've had years of demos showing how our system can build an app for a customer in under a minute, but we rarely use it because we don't think customers will believe it."
This is no longer the case. With interest in generative AI so strong, one-minute demos have become "at the heart of our conversations with customers," and the company has adjusted its marketing and product roadmaps to emphasize its generative AI capabilities.
"The way the world thinks has changed, and the market suddenly has a thirst for what we've been building," Kulkarni said. "When you have moments like these, you need to meet them."
This article is a compilation article for the deep thinking circle, please contact the public account background for reprinting
Reference materials
[1]https://www.nea.com/blog/4-trends-for-ai-startups-and-generative-ai-companies
end
The main content and views of this article come from the founders of NEA's four invested companies in the field of generative AI, so naturally there is also a PR component in it. But these four companies are currently very good benchmarks for the development of generative AI, so many of the cognitions and opinions are also worth carefully scrutinizing and absorbing the essence. For more analysis of plugins and AI ecosystems, you can refer to my previous article - AGI is coming? New phases and trends in AI Paradigm.