AI Agents Do More Than Answer Queries --- Autonomous AI can perform complex tasks, but still must prove its utility

"The spotlight in artificial intelligence is moving from chatbots to so-called agents, autonomous AI that can follow complex instructions and perform tasks from checking a car-rental reservation at the airport to screening sales leads.

The first questions about AI agents as they enter the market will focus on how capable the new systems are, including at making decisions with little or no human guidance. If the technology proves itself, they could go a long way toward addressing businesses' frustrations as they try to recoup investments in AI. But other questions will follow.

"It's not clear to me these things will ever effectively stand alone, but the instructions that you will be able to give them will be increasingly high-level," says Martin Casado, a general partner at venture-capital and investment firm Andreessen Horowitz, where he leads the firm's $1.25 billion infrastructure practice.

Big players are betting on agents for a range of uses. An Apple executive used the company's developers' conference in June to paint the picture, describing the ways an AI-boosted Siri might help during a hypothetical visit from her mom by delving into the executive's email, messages and other apps with minimal prompting.

Dating app Grindr says it is working on an AI-agent wingman that will recommend relationship candidates, help users keep track of conversations, suggest date spots and make restaurant reservations.

Cloud software provider ServiceNow said in September it plans to integrate AI agents into its platform that automates and manages functions such as IT, customer service, procurement, human resources and software development. The first piece ServiceNow is shipping includes the autonomous handling of IT and customer-service-management ticket requests, said Dorit Zilbershot, vice president of platform and AI innovation at ServiceNow.

"We're moving into a world where. . .I oversee the AI, the AI is able to handle all those tickets, and they reach out to me when they need help or they reach out to me when, as part of the policy, they have to get my approval," Zilbershot said.

Data-analytics company Palantir Technologies, which was added to the S&P 500 in September, helps clients develop their own agents and also uses them internally for functions such as legal work, human resources and office operations and facilities.

"The places where AI is driving differentiated value go beyond chatbots. Chatbots are table stakes," Palantir Chief Information Officer Jim Siders said this month at The Wall Street Journal CIO Network Summit. "It happened fast and I think. . . .a lot of the industry is catching up to that."

The former co-chief executive of software-as-a-service pioneer Salesforce, Bret Taylor, has started a company called Sierra to help other companies build agents.

Salesforce this summer itself introduced an offering called Agentforce, which it says can automate customer service, marketing campaigns and business procurement far beyond any chatbot's abilities.

Companies can use Agentforce not only to prequalify business leads, for example, but to then reach out to a promising prospect and set up a meeting on behalf of a human salesperson, according to Salesforce.

BACA Systems, a machine manufacturing company in Orion Township, Mich., began using Agentforce capabilities nearly a year ago, when they were in the pilot stage.

The technology has provided a boost to BACA by enabling the company, for example, to handle a growing volume of service calls without adding additional human staff, according to Andrew Russo, enterprise architect at BACA.

Companies are already familiar with other forms of automation, which will help lay the groundwork for trying AI agents, according to research and advisory firm Forrester.

But they will run into challenges, too, because the norms over automation that they have evolved won't translate neatly to the implementation of AI agents.

AI agents will force companies to answer new "hard questions" about balancing the risks and rewards of automation, the proper role of humans and the best way to govern the data that these increasingly powerful agents use, Forrester said in a report this month.

AI providers and their clients may at least be more aligned than before on pricing and incentives. Providers of generative AI tech have often charged on the software-as-a-service (SaaS) model, sometimes pegged to the number of employees with access to the tech.

SaaS in general has been a welcome improvement over previous licensing models that often took many years and billions of dollars for customers to implement. But it isn't perfect. It's possible to oversubscribe -- and pay -- for services companies don't wind up using.

Instead of a continuing fee, Salesforce is charging $2 per conversation or sales lead for Agentforce.

That may help attract customers that wouldn't have been ready to sign up for AI agents if it meant a new recurring fee for a service that might not turn out to be valuable enough. For Salesforce, the model creates an incentive to deliver clear value.

The SaaS pricing model won't go away, but the newer alternative makes sense because work tied to agents isn't always linked to an individual human employee, according to Adam Evans, senior vice president of Salesforce AI Platform.

Agents will be able to perform certain kinds of work that humans simply don't have the time to give priority to, such as analyzing all the calls related to an area such as logistics, he said.

"We're teeing this with consumption so that every customer is going to be able to say what job did these agents do, see their bill almost by the job or the campaign," Evans said. "And then they can do their own [return on investment] calculation and drive those decisions they should be making."" [1]

1. AI Agents Do More Than Answer Queries --- Autonomous AI can perform complex tasks, but still must prove its utility. Rosenbush, Steven.  Wall Street Journal, Eastern edition; New York, N.Y.. 21 Oct 2024: B.4.