{"text":[[{"start":5.6,"text":"How do workers react when an AI agent joins the group chat? Cautiously, according to Yoav Levy, co-founder and chief executive of cyber security company Upstream Security. "}],[{"start":17.15,"text":"After he added an agent to his marketing team’s Slack channel, there were “24 hours when no one wrote anything”, he says. “People were afraid. Then we had the first brave person writing “‘Hey!’ — and the bot starts having a chat.”"}],[{"start":30.349999999999998,"text":"Caution is an understandable human reaction. While large language models (LLMs) answer questions and perform tasks on command, agents have, well, agency. “Agentic AI is AI that can plan and execute complex workflows and can plan into the future,” explains Nicholas Wright, a neuroscientist who researches the brain, technology and security at University College London and Georgetown University. "}],[{"start":58.5,"text":"Most businesses are still in the early stages of working out how best to integrate AI agents into their teams and work processes. As they do, corporate leaders need to consider the potential productivity gains — but also the effects on staff. "}],[{"start":73.25,"text":"AI agents are still more talked about than deployed: outside the technology sector, rollout is “jagged”, according to Stephen Chase, global head of AI and digital innovation at KPMG. Just 9 per cent of businesses say they have active agents across different parts of their organisation, according to KPMG’s latest Global AI Pulse report. Around a fifth are “exploring the possibility” of using agents, but have not yet implemented them, and 17 per cent are running pilots."}],[{"start":99.7,"text":"When Chase talks to corporate bosses, there is often a “general sense that it’s OK to be a fast follower” in this area, rather than aiming to pull ahead of competitors."}],[{"start":108.95,"text":"Companies where agents are already well embedded can provide some lessons on what such a process involves."}],[{"start":115.7,"text":"Israel-based Upstream, which employs about 150 people, found that its first agent in the marketing Slack chat immediately started to boost the team’s efficiency. “We started to see it can identify opportunities that are stalled, lost [or] people forgot about them,” says Levy. “Then we moved group by group, and started using them internally.” "}],[{"start":135.65,"text":"The agents are used for routine jobs, and “the sweet spot” for Upstream is having them run what were previously time-consuming tasks, where data has to be extracted from multiple sources."}],[{"start":146.75,"text":"One example is for clients in the mobility sector, where staff work with agents to stop criminals from pairing with in-car systems and apps in order to hack them. Agents can also investigate whether a fault is restricted to one car, or whether there is something systematic affecting all models from a particular year, for example. “The agents are helping humans to conduct this investigation, doing triage, looking at multiple tools to correlate everything they see within minutes,” Levy says. “The humans who are working with the agents in the same Slack group can go to the root cause analysis much faster.”"}],[{"start":181.65,"text":"Adecco Group, one of the world’s largest recruitment and human resources companies, has about 35,000 staff globally and is rolling out agents to fit into its plan to have 50 per cent of revenues “powered by agentic AI” by the end of this year."}],[{"start":196.05,"text":"The company started trialling agents a year ago in its UK recruitment business. Niki Turner-Harding, UK and Ireland country head at Adecco, says the team began by analysing every step of its recruitment process, to understand “where an agent might add value, and where a human is essential”."}],[{"start":214.75,"text":"Their initial analysis identified pre-screening for job candidates as a task that could be handed to AI. The resulting agent operates 24/7, unlike human recruiters, who keep office hours. Turner-Harding says she and her team were initially “shocked” to find that 50 per cent of candidates’ conversations with the agent took place outside the typical working day, peaking between 11pm and 3am. “It meant that previously we were missing out on 50 per cent of candidates,” she says. Time saved in the recruitment process, so far, is 20 per cent. The company says it has focused extra staff time on coaching and supporting candidates, rather than reduced headcount. "}],[{"start":256,"text":"One of the biggest barriers to an agentic rollout is cost. Most software platforms charge clients for using agents through “tokens”, which represent a metered amount of AI thinking capacity. (A recent Deloitte report calls tokens “the new currency” of the AI economy.) Because agents are autonomous (and it is easy to forget about them), staff can run up huge bills for tokens if the business is on a pay-as-you-go contract. Adecco’s contract with Salesforce offers staff unlimited access to agentic AI; an approach that keeps costs in check, and is likely to become the norm for businesses that can afford it. "}],[{"start":292.25,"text":"Prosus, a company that has already built 50,000 agents, may offer a glimpse of the future of work. The AI and ecommerce group, which employs about 23,000 people, operates in several sectors, including food delivery (it owns Just Eat). It also has a stake in Tencent, the Chinese tech group."}],[{"start":310.9,"text":"Prosus’s approach has been to encourage non-technical staff to build and use their own agents using a platform called Toqan. The tool works in a conversational, non-technical way. Euro Beinat, head of AI at Prosus, says: “You log in, and describe the task you want to do.” "}],[{"start":328.04999999999995,"text":"Beinat acknowledges that only about 5,000 agents are in daily use — but building them is part of the learning process for staff. He describes the shift to agentic workflows over the past 18 months not as a technical change, but a cultural one: “It’s a way that an entire organisation thinks about making their work better, faster — whatever criteria you want.” "}],[{"start":349.24999999999994,"text":"Its agents have led to increased independence for staff: legal teams, for example, can create their own data analysts, rather than having to rely on others for advice. In the food delivery businesses, account managers previously spent many hours preparing for restaurant visits, including asking a data analyst to pull sales data. Now they can get reports at the click of a button. "}],[{"start":370.29999999999995,"text":"Beinat estimates agents have so far created efficiency gains equivalent to more than 1,000 full-time employees, although this is hard to measure. He says what is easy to prove is that agents are magnifying the productivity of the company’s already top-performing staff. "}],[{"start":386.15,"text":"But this creates more work for others. The implication is that organisations are going to have to rethink how they structure teams and processes. Beinat says: “If I create 20 times the work, I need 20 people to check what I am doing. So, we are starting to scratch the surface — we need to change the way we work.”"}],[{"start":405.54999999999995,"text":"Companies such as Prosus are still outliers. As Chase at KPMG says, some business leaders are talking about agentic AI but still have not had personal experience with it: “It would be like trying to tell your kid how to ride a bike, but you’ve never ridden one.”"}],[{"start":428.69999999999993,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1777900019_8941.mp3"}