The rise of the AI employee
How would your business change if it was run by 1000 AI employees for the cost of a single entry-level position?
Sequoia Capital recently wrapped up its AI Ascent summit, where they brought in AI leaders from the most successful early-stage startups and public companies around the world to discuss the future of the industry. Even esteemed AI leaders aren’t too sure how the current wave of technology will play out. Although they all agree on one thing; generative AI agents represent the next huge sea change in technology on the scale of mobile phones and the internet. In this article, we’ll zoom in on a few talks that our team found particularly illuminating which struck a chord with our customers (read more here to learn about agents in more detail).
Sequoia’s opening keynote set the stage for how they perceive the next 10 years will play out for businesses that make AI a cornerstone of their operation. They predict that traditionally capital-intensive incumbents in industries like finance and healthcare will leverage agents to massively reduce marginal costs. This would produce a downward pressure on prices to bring them in line with products in computing and software which were one of the only industries that has gotten cheaper for the consumer over time. New entrants to the market that are building AI-based businesses in these areas will have a headstart without the need to migrate existing systems to the “agentic” way of doing things. They express that the true potential is only unlocked when these agents are directly linked to business KPIs that allow them to drive the status quo of a business by incorporating them into their feedback loop for completely autonomous, iterative improvement (read more here if you want to learn what an agent is).
The overall theme is clear, a confluence of events is forcing businesses to adapt and change to this new technological climate, the first wave was LLMs and the next will be agents. A business organization that’s operated by 1000 agents never has to worry about the increasing complexity of communication and alignment that a traditional business has to deal with as it scales operations. More often than not, the greatest obstacle to innovation in established businesses is figuring out where your data is and then finding opportunities for innovation. With agents, this would never be a problem as they always have all the data they need; they can instantly offer insights about areas of the business that could be improved or optimized.
Can AI agent technology make the jump to enterprise?
This poses an immediate question: Is this even possible? CJ Desai, COO of the now famous enterprise automation platform ServiceNow made it clear that AI has already been making an impact for his customers. As a big supporter of open-source LLMs, they’ve put them to work by massively reducing the time it takes to complete tasks like inventory management or invoice reconciliation. For large enterprises, open source models are much more capital efficient which can make a big difference if a 1% change in margin can make or break your quarter. He sings to the same tune as Andrej Kaparthy (Director of AI at Tesla), who agreed that the open source ecosystem has vast amounts of untapped potential, but he still thinks there’s significant room for improvement: “You build a product, then a ramp. But [people] are just building products with no ramp!” he exclaimed. A product with no ramp draws on images of computers before the Apple Lisa. Without a simple point-and-click graphical user interface, computers at the time were relegated to purely technical audiences, despite the huge uplift in productivity they came to embody in the years to come. The added layer of accessibility that Apple introduced suddenly turned the computer into a savvy tool anyone could start with and ramp up to more advanced use cases.
Andrej goes on to point out a big gap between the capabilities AI agents have showcased in tech demos and agents that can produce a real-world impact on your business. For example: creating content that’s consistent with your brand voice to drive engagement across your go-to-market channels. Consistency is critical when it comes to operating agents for these use cases at scale. Fortunately, this technology has been evolving at an extremely rapid pace, and guardrails for this consistency have already been engineered and put to work. One of the greatest advancements here is agent planning, which effectively allows agents to document the actions they intend to take so that they can be corrected and or approved.
From LLMs to AI agents
“For those who haven’t experienced agents that plan [ahead], it’s a real aha moment on the level of that first experience with ChatGPT,” remarks Andrew Eg, co-founder of Coursera and Professor of Machine Learning at Stanford. He echoed the message in the opening keynote: As AI agents can automate more and more of the workplace, we’ll begin to see entire organizations that just operate as a unified team of agents, constantly working to optimize and improve the KPIs that drive your business. It remains to be seen if the technology we use today will become established as the go-to method or if we’ll invent new ones along the way. What’s for certain is that businesses need to start taking a serious look at how AI will impact their strategy in the next 12 months and start finding the easiest ways to test the waters; that’s where Solitude has you covered.
We want to make running 1000s of agents for your business simpler than your morning email routine. As new AI use cases emerge and agents become commonplace, managing them and orchestrating their work will become astronomically difficult, especially if their services and billing are spread out across 10–20 different vendors. If Sequoia’s vision plays out, this will be the least of the problems when dealing with agents sprawled across so many different platforms. What about governance and security? You shouldn’t have to worry about all this as running your business is already hard enough. We believe that managing and orchestrating fleets of agents should be as easy and intuitive as any other software you use today; point, click, complete. Other leaders in the space seem to agree with us too, all of whom reference the promising results that enterprises have been achieving today while acknowledging the shortfall in the UX of existing products.
“UX for agents is underbaked” Harrison Chase sighs. As one of the creators of Langchain, a technology that underpins agents, he agrees that AI agents are here to stay and will change the way that we work even though they present a serious challenge in terms of usability. “Chat boxes are insufficient” and if you’re an enterprise trying to solve real-world problems then the truth of this statement is immediately apparent. We too discovered that chat boxes are not suitable for scaling enterprise workflows because human users need to have insight and control of the actions the agents are taking. As a decision maker, understanding the thought process and reasoning that went into any content or output the agent generates before it is actioned is critical. Agent notifications and human approval workflows keep you constantly informed of key events and outcomes that the agent was responsible for while ensuring you have the final say on any action the agent pursues. For our customers, access to a central dashboard where they can freely view and approve any requested actions by agents is critical. This helped them avoid the flood of emails and notifications from other systems that can quickly become white noise.
The future of agents for the enterprise is extremely promising when you look at the vast range of use cases they can deliver across marketing, sales, customer service, and even software development and finance. With 1000s of agents at your fingertips, 1-person businesses that breach $100m+ ARR are not too far off. The technology is here today, but key issues around management and governance must be addressed first before we can confidently put them into production, which is why we built Solitude with those core principles in mind.
If you’re curious for a sneak peek at some of the platform features we mentioned today, our website is a great place to start: https://www.solitude.ai/
Sources
- Opening Keynote
- https://www.youtube.com/@sequoiacapital
- Harrison Chase
- https://www.youtube.com/watch?v=pBBe1pk8hf4&ab_channel=SequoiaCapital
- Andrew Ng
- https://www.youtube.com/watch?v=sal78ACtGTc&ab_channel=SequoiaCapital
- Andrej Karpathy
- https://www.youtube.com/watch?v=c3b-JASoPi0&ab_channel=SequoiaCapital
- CJ Desai
- https://www.youtube.com/watch?v=-hUJTEDOQDs&ab_channel=SequoiaCapital
