Authors: Jake Nibley | Vice President, Tola Capital and Nischal Nadhamuni | CTO and founder of Klarity
Author Bios: Jake Nibley is Vice President at Tola Capital, a venture capital firm that focuses on the once-in-a-generation opportunity to invest in the next wave of enterprise software companies enabled by the new data and cloud platforms. As an investor, he applies this experience using deep financial forensics to understand the underlying health and drivers of businesses and to identify strategic growth opportunities.
Nischal Nadhamuni is the CTO and founder of Klarity, a company on a mission to automate document centric workflows. An MIT grad (class of 2018), Nadhamuni developed machine learning expertise and experience as an undergraduate via his classes and internships, including at Flipkart, where he created an automated tool to detect fraudulent user behavior. After meeting Andrew Antos in an entrepreneurship class at MIT in 2016, an idea sparked: automating document review would simplify workflows for millions of workers around the world. In 2017, the two formally launched Klarity.
A decade ago, Andreessen Horowitz famously proclaimed that ‘software is eating the world.’ At this moment, it’s safe to say that AI is the mouth. Generative AI is changing how we think about the speed and scale of software innovation. So much so that it has the potential to change the economy. According to PricewaterhouseCoopers, AI could add up to $15.7 trillion in global GDP in 2030, more than the current output from India and China combined.
Companies big and small are clambering to find ways to leverage generative AI in their products. When OpenAI launched GPT-4, they initially partnered with a handful of select companies, such as Intercom, Stripe and Duolingo, to showcase GPT-4’s capabilities. But the floodgates are open with API accessibility making GPT-4 more broadly available to companies of all sizes.
Now is the time for founders and entrepreneurs to lead their organization’s adoption of generative AI and LLMs like GPT-4. Because this development is happening so rapidly, companies must match that same agility and speed while innovating to stay competitive. That’s why you need to take a generative AI sabbatical. This defined reset will force you to radically rethink what’s possible once generative AI becomes mainstream.
A generative AI sabbatical in action
When companies consider integrating generative AI into their products, we rarely see AI-native user interfaces. Instead, many are making the mistake of pursuing incremental tactical wins to be a “Gen AI company,” like adding a ChatBot to their customer service experience. Companies and founders should think bigger about the capabilities that were never fundamentally part of their company’s vision but are now possible with generative AI lifting previous technology constraints. That mind shift is the ultimate value of taking a generative AI sabbatical.
We know how powerful the practice of a generative AI sabbatical is because we’ve seen it firsthand with our portfolio company, Klarity. At Klarity, AI is the foundation of their document review automation platform. When OpenAI released ChatGPT 3.5, they saw the scale of what it was capable of and were immediately curious about how to leverage it. The experience pushed the company to rethink some of our preconceived notions about how generative AI could improve their product and what their company could be capable of in the future.
Klarity’s CTO, Nischal Nadhamuni, decided to put all other new development work on hold for a four-week hiatus and maintain a skeleton crew for code maintenance. They called this their Generative AI Sabbatical. Just like during an academic or research sabbatical, they wanted to hyperfocus on this research project of leveraging GPT3.5 to improve the customer experience. They discovered that it completely changed how they thought about shipping code, the pace of innovation, and what’s possible with LLMs and generative AI.
A generative AI sabbatical roadmap for founders
Any company can test this practice for themselves, whether they’re a large organization with limitless resources or a small, scrappy startup. The key is balancing the vision and impact of generative AI without isolating existing customers. Building features that prioritize the greatest good for customers and your business’s longevity may force you to make short-term resource tradeoffs. But the outcome may be worth it.
Nischal and his team at Klarity created a roadmap to follow if you consider piloting a generative AI sabbatical. These key components include setting experiment parameters that lead with the company’s vision, creating a team structure, rigorous hypothesis testing, and implementing the results into the company’s ethos.
1. Identifying experiments: Lead with the company vision
To identify the success or failure of the experiment, you need to know what you’re trying to measure. Because the experiment is designed for a short period of time, narrow the scope by creating a hypothesis you want to test.
Start by recognizing this is not a technical exercise; it’s about the company vision, so the entire core leadership must be involved. Nischal began with an educational session with the Klarity leadership team on generative AI, its history, and the latest breakthroughs in the technology. Before the next session, each executive had homework; try ChatGPT on various personal and professional tasks.
When they reconvened in the morning, they shifted focus to Klarity’s company vision. The prompt was simple: how will generative AI change the company’s vision? “For 3 hours, we huddled around our conference room table and threw dozens of ideas onto a Miro board,” Nischal told me. “The only constraint: radical ideas only, nothing incremental.” Klarity’s founders then distilled these ideas into five central ideas or capabilities that generative AI would need to satisfy.
The Klarity team compartmentalized this work into five distinct workflows and worked through the experimentation process with each. The sabbatical was designed to last one month, but they finished in three weeks because they knew what we were testing for and had clear guardrails in place before the experiment began.
2. Team structure
Depending on the size of your organization or the level of engineering work required to keep your product’s lights on during the experiment, it may not make sense to isolate your entire engineering team. You might be able to split your teams into designated work streams — like a blue and a green team — where one team (green) will keep things running as-is, and the other (blue) team is using all their time to experiment for a set length of time.
3. Execution
With the parameters of your experiments defined and your special teams built out, it’s time to execute. Nischal’s team explicitly named the four ways this type of experiment would have to differ from traditional product building to succeed.
- Speed: This would be an all-out sprint unlike anything else. Cycle times would be in hours, not days or weeks. Fundamentally, generative AI models are so generalizable and so easy to interact with that they allow for this kind of velocity.
- Learnings over product: Engineering had full license to write scrappy inelegant code and ignore downstream deployment challenges; those would come later. Success would be measured on whether they executed the experiments successfully — product building was secondary. Because of this, falsifications and negative results were equally as valuable as positive ones.
- Radical ownership with radical collaboration: Each experiment was assigned to one engineer and one engineer only. Because there were so many shared areas of learning, collaboration was not only heavily encouraged, it was essential. We set up daily meetings to discuss all workstreams and weekly Friday meetings to share a roundup of all the breakthroughs in generative AI from the last week.
- Leadership involvement: There is no substitute for technical leadership being intimately involved in the entire experimental process, including at the code level. Developing generative AI intuition at the highest level is vital since this has the potential to change every aspect of product building.
4. Looking ahead to the post-experiment future
This moment in technology is unlike any we’ve seen since the invention of cloud computing. And since consumers regularly interact with generative AI with ChatGPT or Bing, their expectations for how this technology will be integrated into other products are changing. The level of speed and creativity needed to meet the moment means that leaders must lead differently.
After Klarity’s sabbatical, Nischal and his team began meeting more frequently but for a shorter amount of time. This allowed them to quickly check in and share expectations around deliverables — even down to the hour that they’ll be delivered — because the technology was changing so rapidly. They also shifted their team structures to be smaller and more cross-functional, which gives individual contributors much more ownership over their work. The Klarity team has engineers who own entire features, acting effectively as product managers — a strategy that OpenAI founder Sam Altman talks about often.
As we move into a space where generative AI research and development becomes part of our daily work, it’s even more important to stay focused and not fall down AI information rabbit holes. This could come in the form of Slack groups focused on sharing ideas or reading groups where members share the AI-focused articles and reports they’re reading that week. It’s essential to recognize that the vast majority of generative AI that will be implemented into a company’s product has yet to be invented. What we’ve seen today is just the tip of a massive iceberg, so companies and teams have to become adept at keeping their eyes and ears open to the developments of tomorrow while staying on top of shipping code today.
Will this generative AI wave be a headwind or a tailwind?
Brad Porter, the former CTO of Scale AI, recommends that founders think about building their products so that as AI improves, it creates tailwinds instead of headwinds. For some companies, generative AI will be a tailwind, as it is for Klarity. Klarity is using AI to automate document-centric workflows, and generative AI dramatically accelerates their time to achieve their company vision.
But for other businesses, this wave could quickly become a headwind, an existential threat to their vision. Many companies could become the next Blockbuster, with their Netflix-esque competition making their product obsolete because they didn’t innovate quickly enough to embrace generative AI. The sooner founders understand that their vision is either validated or at risk because of generative AI, the better they can pivot and ride the wave. The generative AI sabbatical is one way companies and tech leaders can innovate faster and challenge their preconceived notions of what their vision could be.