The AI Daily Brief: Artificial Intelligence News and Analysis - Why AI Agents Will Do for the Enterprise What RPA Never Could

Episode Date: November 17, 2024

A reading and discussion inspired by https://a16z.com/rip-to-rpa-the-rise-of-intelligent-automation/ Brought to you by: Vanta - Simplify compliance - ⁠⁠⁠⁠⁠⁠⁠⁠⁠vanta.com/nlw⁠⁠⁠⁠⁠⁠⁠⁠⁠The AI Daily Brief helps you understand the most important news and discussions in AI. Subscribe to the podcast version of The AI Daily Brief wherever you listen: https://pod.link/1680633614 Subscribe to the newsletter: https://aidailybrief.beehiiv.com/ Join our Discord: https://bit.ly/aibreakdown

Today on the AI Daily Brief, a look at how agents might fill the potential once seen in RPA. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes. Hello, friends, happy weekend. We have a Long Reads episode for you here today. In my conversations with enterprises, one of the things that they are watching most closely is where automation solutions in the form of agents are really going to actually be performant in a way that allows them to replace entire categories of tasks. This is a road that many organizations and enterprises have been down before

in the form of something called robotic process automation or RPA. RPA in practice has had a very spotty record of success. In the piece that we are reading today, or rather that I am turning over to AI to read, Kimberly Tan of Andresen Horowitz explores how agents are replacing RPA and living up to the original promise. So let's read the essay, and then we'll come back and do a little bit of a discussion.

As AI turns labor into software, the opportunity to productize external professional services, for example, in legal or accounting, has become a hot topic. However, we believe there is also substantial opportunity in productizing internal work within organizations. These responsibilities often fall under the umbrella term of operations and can range from full-time data entry and front desk roles to routine operational tasks embedded in every other role. This work generates fewer media headlines, but it is the internal stitching that holds companies together. These ops roles involve critical, but often repetitive and mundane tasks. Companies have historically attempted to automate these tasks by using robotic process automation, RPA,

but with generative AI, we believe true automation through agents is now possible. We've already seen early examples of agents working in production, such as Decagon's automated customer support. And with companies like Anthropic launching capabilities like computer use to enable models to meaningfully interact with existing software, there is a clear emerging infrastructure stack for founders to build verticalized intelligent automation applications. These examples preview a world in which AI agents are able to fulfill the original promise of RPA, turning what used to be operations headcount into intelligent automation and freeing workers to focus on more strategic work. The original promise of RPA and the impact of AI. Operations work is sprawling and diverse, including tasks like data entry, document extraction,

information transfer, system migrations, and web scraping. These tasks are essential, but they often lack the APIs or direct integrations required for traditional software to manage them efficiently. Tons of work is still done over phone calls, spreadsheets, fax lines, and paper forms. And over the last decade, RPA became a buzzword for automating this type of work. Companies like UiPath, which was founded in 2005, promised to enable the fully automated enterprise and empower workers through automation. But despite its IPO in 2021 and its current valuation, these last-generation RPA companies couldn't fulfill the promise of true automation. The technology at the time just wasn't advanced enough. As a result, instead of true automation, these companies observed how their customers

navigated a process, then built bots that mimicked the exact keystrokes and clicks that a human would make. While these bots often provided meaningful business value when they functioned correctly, they stumbled if the process was not rigid and clearly defined, or when it underwent changes. In addition, implementing these bots required expensive consultants, which meant RPA was only available to companies large enough to afford this heavy-handed approach. With LLMs, however, we believe the original vision of RPA is now possible. Instead of hard-coding each deterministic step in a process, AI agents will instead be prompted with an end goal. For example, book an appointment for the customer, transfer data from this document into this database, and then be empowered with the right tooling and

context to take those actions on behalf of the company. They'll be adaptable to various data inputs and capable of handling changes in business processes, and because of this flexibility, they will be far easier to implement and maintain than traditional RPA systems. The future of AI ops and where the opportunity lies. We're excited about this opportunity in intelligent automation for two main reasons. The potential market is enormous. For all the work that current software can handle, there are orders of magnitude more work that it cannot. Work that is being done via pen and paper, spreadsheets, phone calls, and facts. Intelligent automation can address the current labor costs associated with this work, comprising over 8 million operations,

information clerk roles according to the Bureau of Labor Statistics, as well as the spend associated with outsourcing this work, representing a meaningful portion of the $250 billion business process outsourcing. Startups largely have a greenfield opportunity in this space. There is often no existing software product for these workflows given their bespoke nature. The people were the product. As a result, these roles never developed systems of record in the way other roles did. For example, Salesforce for Sales, Workday for HR, meaning there is no software incumbent to add AI into their existing product suite. This market is wide open for startups. Specifically, we view the market opportunity as focused on two main areas, horizontal AI enablers that execute a specific function

for a broad range of industries, and vertical automation solutions that build end to end workflows tailored to specific industries. Today's episode is brought to you by Vanta. Whether you're starting or scaling your company's security program, demonstrating top-notch security practices, and establishing trust is more important than ever. Vanta automates compliance for ISO-2-GDPR and leading AI frameworks like ISO-42,1, and NIST AI Risk Management Framework,

saving you time and money while helping you build customer trust. Plus, you can streamline security reviews by automating questionnaires and demonstrating your security posture with a customer-facing trust center all powered by Vanta AI. Over 8,000 global companies like Langchain, Lila AI, and factory AI use Vanta to demonstrate AI trust and prove security in real time. Learn more at vanta.com slash NLW. That's vanta.com slash NLW. Today's episode is brought to you, as always, by Superintelligent.

Have you ever wanted an AI Daily Brief, but totally focused on how AI relates to your company? is your company struggling with AI adoption, either because you're getting stalled, figuring out what use cases will drive value, or because the AI transformation that is happening is siloated individual teams, departments, and employees and not able to change the company as a whole? Super Intelligence has developed a new custom internal podcast product

that inspires your teams by sharing the best AI use cases from inside and outside your company. Think of it as an AI Daily Brief, but just for your company's AI use cases. If you'd like to learn more, go to B-Super.a-I-SlarHplner and fill out the information request form. I am really excited about this product, so I will personally get right back to you. Again, that's B-Super.a-I. slash partner. Horizontal AI enablers.

Today, every intelligent automation company is building a similar set of capabilities and internal tooling. This creates a perfect opportunity for startups to simplify the process by focusing on one specific foundational component. For example, almost every intelligent automation company has to parse unstructured data and output contextualized structured data. Many companies have built this out internally and companies like Reducto and Extend are working to be the horizontal enabler to solve this specific need. We think there are many other core building blocks needed for complex intelligent automation, including but not limited to building web data crawlers, structuring data from unstructured sources, or writing data back to legacy systems. end-to-end vertical automation. We've previously written about our excitement for investing in vertical software, software that sells to one particular industry. We think this is a particularly good fit for intelligent automation since operational agents will need to have the narrower context and deep

integrations to achieve the accuracy and consistency customers expect. Every industry has back-office operations that could be automated, and we've already seen startups use LLMs to automate one flow as a strategic wedge to build deeper for specific industry needs. In health care, for example, Tenor has automated the referral management flow. Referrals are the lifeblood of any growing health care practice, but accepting a referral used to require a lot of manual labor, for example, receiving a fax, having the front desk pull the information from the fax and manually inputting that patient information into their system. Tenar has built intelligent automation to solve this information transfer problem, using LLMs to extract unstructured data from PDFs and faxes, run valid,

on the information and then write that information back into the system of record, EHR, automatically. This dramatically reduces the time it takes to accept a referral, which allows customers to secure new business more quickly. In logistics, trucking brokers spend an enormous amount of time processing inbound orders and tracking loads. Now using intelligent automation, companies like Happy Robot can automatically check on load status and updates via AI-powered voice assistance, and companies like Vuma are able to ingest unstructured email data to automate price quoting and order entry into the trucking management system, TMS. These companies often focus on automating a very narrow but very common and important

workflow in their respective industries, often involving data and information transfer. They do not seek to be the system of record, at least initially, and can thus bypass the difficult rip-and-replace problems of going after legacy systems. They also start by automating revenue-generating workflows making themselves top priorities for their customers. Because these automation start at the beginning of a workflow, these startups earn the right to the upfront data and downstream workflows. We believe this approach is a winning formula for intelligent automation startups, and we're

eager to partner with those going after this opportunity across different industries. We are incredibly excited by the future of intelligent automation. LLMs have given startups the opportunity to fulfill the initial promise of RPA. By automating tasks traditionally handled by labor, they can now tap into markets and opportunities that were previously too small or too difficult to pursue. We believe a number of large companies will be built here, both in the horizontal enabling layer and in the verticalized end-to-end solution

for customers in different industries. All right, back to Real Non-AI-NLW here. A couple things that are really interesting about this paper to me. The first of all is just the opportunity in general. One of the things we discuss all the time with artificial intelligence is the idea that we're not looking to replace humans entirely, we're looking to take off their plates tasks that are extremely repetitive and mundane. Now, of course, if we are being honest with ourselves, inevitably, there will be some categories

of jobs that do so many of those tasks that they will be disrupted. But by and large, the goal, again, isn't to disrupt people. It's to make work that is repetitive and mundane but necessary, automatic and automated. The fact that there is so much fertile exploration of agents who are focusing in a vertical way on that set of tasks, I think is really promising. Now, speaking of vertical, one of the really interesting games, apps in the AI space is the space between those who are trying to create generalist agents, which include both the frontier labs as well as some startups like Multion, and those who are

thinking instead about extremely specific vertical agentic applications. Now, the risk for the vertical companies is, of course, that the frontier labs figure out something and their generalist agents are just more performant even than the vertical agents are. However, I think that from where I'm sitting, the most likely scenario for how agents actually come to market is going to be in these highly verticalized applications. Now, mostly when people talk about vertical applications, they're talking about an industry vertical, I think there will also be functional verticals, specific types of tasks that get done across different industries where agents have the ability to thrive and specialize. Over the next couple years, the pilots that have been focused on

assisted work and co-pilots inside the enterprise are going to increasingly give way to pilots that are focused on agentic. But importantly, despite what Mark Benioff of Salesforce says, that doesn't mean that the assistant era of AI is somehow over. What I believe is that every single business process that we have today is going to be AI-ified in some way. I think that a lot of processes are, yes, going to be automated and taken care of by agents, but I also think a lot of processes are going to be done by AI-assisted superhumans. And what's more when it comes to the really valuable new opportunities, the things that companies can do that simply weren't possible before, I think there's

going to be a ton of human innovation and orchestration. In other words, the big task, for companies over the next couple years is not thinking about who they can replace with robots. It's about how they get the robots and the humans sitting together to be more powerful than either one could be alone. For now that's going to do it for today's AI Daily Brief. Appreciate you listening as always. And until next time, peace.