/cdn.vox-cdn.com/uploads/chorus_asset/file/25362060/STK_414_AI_CHATBOT_R2_CVirginia_C.jpg?w=696&resize=696,0&ssl=1 "Agents are the future that AI companies promise — and desperately need")
/cdn.vox-cdn.com/uploads/chorus_asset/file/25362060/STK_414_AI_CHATBOT_R2_CVirginia_C.jpg?resize=1200,628&ssl=1&description=Agents+are+the+future+that+AI+companies+promise+%E2%80%94+and+desperately+need){kind=link}
People have been automating their tasks for centuries. Now artificial intelligence companies see a path to profit in exploiting our love of efficiency, and they have a name for their solution: agents.
AI agents are autonomous programs that perform tasks, make decisions, and interact with environments with little human intervention. They are currently the focus of every vast company working on artificial intelligence. Microsoft provides “Copilots” designed to lend a hand businesses automate activities such as customer service and administrative tasks. More recently, Google Cloud CEO Thomas Kurian presented an offer for six different agents to increase AI productivityand Google DeepMind just hired an OpenAI co-lead on its AI video productSora, to work on developing simulations for training AI agents. Anthropic has released a feature for its AI chatbot, Claude, that will allow anyone to create their own “AI assistant.” OpenAI includes Level 2 agents in its 5-tier approach to achieving AGI, or human-level artificial intelligence.
Of course, computing is full of autonomous systems. Many people have visited a site with a pop-up customer service bot, used the features of an automated voice assistant like Alexa Skills, or written a humble IFTTT script. But artificial intelligence companies say “agents” – better not call them bots – are different. They believe that instead of following a straightforward, routine set of instructions, agents will be able to interact with their environment, learn from feedback, and make decisions without constant human intervention. They could dynamically manage tasks such as making purchases, booking travel or scheduling meetings, adapting to unforeseen circumstances and interacting with systems that could include humans and other artificial intelligence tools.
Artificial intelligence companies hope that agents will enable monetization of powerful, pricey artificial intelligence models. Venture capital is flowing into AI agent startups that promise to revolutionize the way we interact with technology. Companies expect a leap in productivity, with agents handling everything from customer service to data analysis. For individuals, AI companies are ushering in a novel era of productivity where routine tasks are automated, freeing up time for original and strategic work. The ultimate goal for true believers is to create AI that is a true partner, not just a tool.
“What do you really want” – OpenAI CEO Sam Altman he said MIT Technology Review earlier this year “is this the only thing that isn’t helping you.” Altman described the AI killer app as “a super knowledgeable colleague who knows absolutely everything about my entire life, every email and every conversation I’ve ever had, but doesn’t feel like an extension.” It can solve straightforward tasks immediately, Altman added, and for more complicated ones it will attempt to complete them, but will come back with questions if necessary. Tech companies have been trying to automate the personal assistant since at least the 1970s, and now they promise they’re finally close.
At OpenAI’s press conference ahead of its annual Developer Day, Director of Developer Experience Romain Huet demonstrated Realtime’s novel API alongside an agent assistant. Huet gave the agent a budget and certain restrictions for purchasing 400 chocolate-covered strawberries and asked him to place an order by phone at a fictitious store.
The service resembles Google’s booking bot called Duplex from 2018. However, this bot only performed well in the simplest scenarios – it turned out that a quarter of its calls were actually made by humans.
Although the order was placed in English, Huet told me he gave a more complicated demonstration in Tokyo: he asked the agent to book him a hotel room in Japanese, where he would conduct a conversation in Japanese, and then call him back in English to confirm it done. “I obviously don’t understand the Japanese part – I just get by with it,” Huet said.
But Huet’s demo immediately caused concern in a room full of journalists. Can’t AI assistant be used for spam calls? Why didn’t it present itself as an artificial intelligence system? (Huet updated the demo for the official Creator Day, an attendee says, leading the agent to identify himself as “Romain’s AI Assistant.”) The anxiety was palpable and not surprising — even without agents, AI tools exist already used for fraud.
There was another, probably more immediate problem: the demo didn’t work. The agent lacked sufficient information and misregistered the dessert flavors, which resulted in flavors such as vanilla and strawberry being automatically added to the column rather than claiming no such information was available. Agents often encounter problems with multi-step workflows or unexpected scenarios. And they burn more energy than a conventional bot or voice assistant. Their need for significant processing power, especially when inferring or interacting with multiple systems, makes them pricey to operate at scale.
AI agents offer a gigantic step forward potentialbut for everyday tasks they are not yet significantly better than bots, assistants or scripts. OpenAI and other labs aim to improve their reasoning through reinforcement learning hoping Moore’s Law continues to ensure cheaper and more proficient data processing.
If AI agents aren’t very useful yet, why is this idea so popular? In brief: market pressure. These companies leverage powerful but pricey technology and are desperate for any practical applications they can find Also charge users fees. The discrepancy between promises and reality also creates a fascinating hype cycle that drives funding, and it just so happens that OpenAI raised $6.6 billion right after it started promoting agents.
Over the last 12 months, AI agent startups have raised $8.2 billion in investor funding
Massive tech companies have been rushing to integrate all kinds of AI into their products, but they’re hopeful that AI assistants in particular could be the key to unlocking revenue. Huet’s AI calling demo exceeds the capabilities of currently available large-scale models, but he told me he expects such features to appear more frequently as early as next year as OpenAI refines its o1 “reasoning” model.
For now, the concept appears to be mostly hidden in enterprise software stacks rather than consumer-facing products. Salesforce, a provider of customer relationship management (CRM) software, launched its “agent” feature a few weeks before its annual Dreamforce conference to great interest. This feature allows customers to exploit natural language to build a customer service chatbot in minutes via Slack, rather than spending a lot of time coding it. Chatbots have access to a company’s CRM data and can process natural language more easily than a bot not based on vast language models, potentially making them better at confined tasks such as asking questions about orders and returns.
AI agent startups (still admittedly a vague term) are becoming quite a fashionable investment. Over the last 12 months, they secured $8.2 billion in investor financing across 156 deals, an 81.4 percent year-over-year enhance. according to PitchBook data. One of the more celebrated projects is Sierra, a customer service agent similar to Salesforce’s latest i project started by former Salesforce co-CEO Bret Taylor. There’s also Harvey, which offers AI agents to lawyers, and TaxGPT, an AI agent that takes care of your taxes.
Despite all the enthusiasm for agents, these risky uses raise a clear question: Can they really be trusted with matters as grave as law or taxes? The AI hallucinations that often plagued ChatGPT users currently have no cure. More fundamentally, how IBM prophetically stated in 1979“a computer can never be held accountable” – and consequently “a computer can never make management decisions.” Rather than autonomous decision-makers, AI assistants are best seen for what they really are: powerful but imperfect tools for low-stakes tasks. Is it worth the gigantic bucks that AI companies hope people will pay?
For now, market pressures prevail, and AI companies are racing to make money. “I think 2025 will be the year that agent systems finally hit the mainstream,” OpenAI’s novel chief product officer Kevin Weil said at a press event. “And if we do it right, we will move to a world where we can spend more time on important human issues and a little less time staring at our phones.”