Rafael Enrique | Light Rocket | Getty Images
Earlier this year, financial services firms Klarna says Its artificial intelligence agents powered by OpenAI have taken over more than two-thirds of customer chats, equivalent to the workload of 700 full-time agents. That’s after just a month of use.
Alexander Kvamme, CEO of customer engagement startup Echo AI, told CNBC that Klarna’s announcement in February could be the first sign that AI agents are “having their ChatGPT moment.”
OpenAI released the ChatGPT chatbot to the public at the end of 2022, allowing the public to experience how the new generative AI chatbot can provide more thorough, creative and conversational answers to online queries compared to traditional search. It’s how consumers seek information online. Google, Microsoft Other companies followed suit with competing products.
The industry quickly moved from text responses to AI-generated photos and videos. Now the rise of artificial intelligence agents.
Agents don’t just provide answers (the realm of chatbots and image generators), but are built for productivity and task completion. They are artificial intelligence tools capable of making decisions, good or bad, “without human intervention,” Kvamme said.
Grace Isford, a partner at venture capital firm Lux Capital, said there has been a “dramatic increase” in interest from technology investors in startups focused on building artificial intelligence agents. Together they have raised hundreds of millions of dollars and seen their valuations climb as the broader generative AI market grows.
According to PitchBook data, generative artificial intelligence grew explosively in 2023, with $29.1 billion invested in nearly 700 deals, and deal value increased by more than 260% from the previous year. Meanwhile, the non-artificial intelligence investment sector has fallen into a lull that has lasted more than two years, following record funding during the coronavirus pandemic.
If 2023 is the peak year of artificial intelligence hype, then 2024 is the year of early deployment.
Jared Spataro, Microsoft’s corporate vice president in charge of artificial intelligence, told CNBC: “Since the launch of ChatGPT, there has been a real torrent of innovation in the market.” Microsoft is the biggest supporter of OpenAI, in addition to providing support to OpenAI. ChatGPT developers have invested billions of dollars in addition to their own generative AI models and products.
The term artificial intelligence agent is not clearly defined in the technology field. Industry experts who spoke to CNBC about this emerging trend generally agree that agents are a step beyond chatbots, as they are often designed for specific business functions and can be customized on large AI models. Think of Jarvis, Tony Stark’s all-purpose artificial intelligence assistant in the Marvel Universe.
Artificial intelligence agents are often described as advanced generative artificial intelligence tools that can perform multi-step, complex tasks on behalf of users and generate their own to-do lists so that users do not have to walk them through the entire process step-by-step.
“The assistant doesn’t just give you an answer, it automates a series of steps,” said Francois Ajenstat, chief product officer at digital analytics company Amplitude.
How Microsoft and Google play
At the Google I/O conference in May, Google announced Project Astra, the company’s latest advancement in artificial intelligence assistants built by Google’s DeepMind artificial intelligence division.
In Google’s demo video, the assistant uses video and audio to help users remember where they put their glasses, check code and answer questions about displayed objects. It’s just a prototype for now, but Alphabet CEO Sundar Pichai said he hopes to roll it out to users later this year.
The demo comes a day after OpenAI showed off a similar message back-and-forth with ChatGPT, positioning it as an artificial intelligence assistant that can act as interlocutor, language translator, math tutor and code co-writer.
Microsoft subsequently announced a partnership with Cognition AI at its Build developer conference, which will bring customers Cognition’s own artificial intelligence agent, Devin. Cognition calls Devin “the first artificial intelligence software engineer.”
Devin quickly became a social media sensation for his ability to handle multi-step processes. Devin doesn’t just produce simple lines of code, but creates a process for solving a problem, writes the code, tests it, and then releases it.
Martin Kon, head of operations at enterprise AI startup Cohere, said AI agents could start doing things like booking and paying for flights, suggesting loan rates, or emailing customers with arrival times and updates. sales force therefore.
Until now, these tools have been largely limited to helping with tasks like writing code. For example, on Microsoft’s GitHub, CEO Thomas Dohmke wrote in a report that approximately 46% of all code “across all programming languages” is generated by artificial intelligence. blog post Early 2023.
While the lines between AI coding tools and true AI agents are blurry, most experts who spoke to CNBC said the defining characteristic of an agent is that it goes well beyond a single use case and starts to approach something like a full-featured personal assistant.
Anthropic and other startups are already working toward this goal. The first step is to enable their chatbots to interact with external tools and services on behalf of customers.
Microsoft’s Spataro said the process of developing the company’s Copilot coding agent was “a bit like being strapped to a rocket.” He said a big part of what Microsoft is doing is moving from one- or two-step tasks to multi-step tasks. This might involve looking at the user’s calendar and taking 30 seconds to look ahead to the day’s priorities.
Fred Havemeyer, Macquarie’s head of U.S. artificial intelligence and software research, wrote in a recent note to investors that the company expects to see more artificial intelligence agents.
“We believe agent AI, which can self-direct tasks, will be the tool to unlock the value of GenAI for everyday users,” Havemeyer wrote.
Romain Huet, director of developer experience at OpenAI, told CNBC that the concept of artificial intelligence agents began to gain traction last year, but people quickly realized that there was still a lot of work to be done to make the tools more autonomous.
“Our models are getting more powerful, so we can now capture user intent better than before, but we’re still in the early stages of building agents,” Hueter said.
The biggest advance, he said, will be for AI agents to learn your preferences and “act on your behalf” without you having to ask.
Startups raise large sums of money
Artificial intelligence agent startups are raking in cash from investors. They are not the multi-billion-dollar-plus rounds that are going into AI model companies, but valuations are still well ahead of business fundamentals.
Adept, led by OpenAI and Google alumni, receives Valuation exceeds $1 billion last year. The company stated that its website Its technology “resolves the complexity of software tools so you don’t have to.”
French artificial intelligence agency startup H raises $220 million seed round In May, investors including Amazon and Samsung UiPath and former Google CEO Eric Schmidt. Artisan AI is a Y Combinator-backed startup dedicated to developing artificial intelligence agents. It calls itself an “enterprise artificial intelligence employee” and recently completed a $7.3 million seed round And said it has cooperated with more than 100 companies so far.
Jaspar Carmichael-Jack, founder and CEO of Artisan AI, said that development of true AI agents will not begin until 2022, when chatbots such as ChatGPT first enable ordinary consumers to interact with such tools.
“People talk about how the venture capital market in general is declining,” Carmichael-Jack said. “But for us, this is like 2021 for AI startups.”
Braden Hancock worked at Facebook Research and the Stanford University Artificial Intelligence Laboratory before co-founding Snorkel AI in 2019. He said it will also take a long time for broader AI agents to become mainstream.
Hancock said agents would have to do “many times better” before people were “willing to accept putting something on autopilot.” He added that when it comes to having technology sign your name and make transfers on your behalf, “the bar is very high.”
Kanjun Qiu’s three-year-old startup Imbue has valuable Backed by the Amazon Alexa Fund and Eric Schmidt, it is valued at over $1 billion. Based on the company’s own user research, Qiu said that the current features of artificial intelligence agents – often intelligent personal assistants that handle delegated tasks – are not what users really want because, by design, they “are not entirely worthwhile.” trust”.
“Even as a CEO, it’s hard for me to delegate things to my executive assistant,” Chiu said. “I’ve had her for two years, and she’s awesome.” As for new things, Chiu said, “It’s still hard for me to fully know, ‘OK, is this going to come back the way I expected it to?'”
Imbue is developing ways for people to run things in the background without coding to meet their individual needs, whether that’s creating a way to track news or building a bot to book travel. These types of AI models do not need to be trained on user profiles because each use case is personalized.
Imbue envisions agents that put control in the hands of the user, rather than delegating tasks to agents built by companies like OpenAI or Google (which centralize and control them).
“There’s one way to think of agencies as enabling everyone to make software,” Qiu said. The user is “asking the agent to write code on the computer to make the computer do what I want.”
watch: “We are still ten years away from solving the problem of artificial intelligence”
About The Author
