5 ways to successfully integrate AI agents into your workplace

Techa Tungateja / Getty Images

ZDNET's key takeaways

• Using AIs effectively means focusing on guidance and governance.
• You'll also need to ensure you train up talent to manage your AI agents.
• Track and trace outcomes to create the perfect tech-enabled work assistant.

It's not unusual to hear business leaders discuss AI being used to augment rather than replace staff, but what does that new working relationship mean in practice?

Also: AI's biggest impact on your workforce is still to come - 3 ways to avoid getting left behind

Five business leaders explain how you can turn AI into a trusted team member.

1. Put clear guidelines in place

Tim Chilton, managing consultant at Ordnance Survey, the UK's national mapping service, said his organization has rules for how professionals can use AI on a day-to-day basis, particularly for generative technologies.

"Copilot is rolled out across our organization," he said. "We have guidelines and training in place for how to use it. I'm in the consultancy team, so it's in our interest to fully understand this technology."

Chilton said he often uses Copilot, with coding and research as the two primary use cases. Other professionals are also getting involved.

"Tools like Copilot are proving useful for desk-based research," he said. "At Ordnance Survey, a large proportion of our staff, compared with other organizations, are involved in research, whether that's through doctorates, university links, or commercial links."

Also: 5 ways to be a great AI agent manager, according to business leaders

More generally, Chilton said the rules and regulations OS has in place are helping staff explore AI-enabled advances in other areas.

"Our data science capability is also investing heavily in model building to benefit us and our customers. We are using these technologies to generate the data we are going to produce in the future and are changing how we capture and quality-assure that information," he said.

"When it comes to geospatial information, we're putting a lot of time and effort into how we use Earth observation data, which is data from satellites, to automatically derive features on the ground, which will then be validated and assured before being put into our models. We're doing that work ourselves and for some commercial customers as well."

2. Have a dialog with your AI

Snowflake co-founder and president Benoît Dageville said we're not yet at the stage where AI acts autonomously in a work scenario, but we might be soon.

"The huge step will be when AIs will really start to do things, such as looking at all your emails and replying for you. Then, you must have full trust, because you are not there to supervise."

However, Dageville also issued a word of warning, suggesting that just because your AI can work independently doesn't mean you should let it.

"Blind trust isn't great either," he said. "It's important to have adult discussions with AI and understand where you need to push the technology and guide it further."

Also: 5 tips for building foundation models for AI

Dageville emphasized the importance of technical considerations and processes, such as governance, security, integrations, and data access permissions.

"AI must not see documents it doesn't have the right to see and shouldn't tell you something that you're not supposed to know. So, the trust is also coming from the platform and all this governance," he said.

"But at the end of the day, AI is like a human being. And as much as you trust someone, you might still listen to what they're saying and say, 'Okay, maybe I half agree with what you're saying, so maybe I should ask another question.' It's a dialog."

3. Develop mid-level talent

Rom Kosla, CIO at Hewlett Packard Enterprise (HPE), said he likes the concept of copilots that act as AI-powered assistants to professionals, and that's something he's explored in his organization.

"We were working with a company that looked at enabling junior development, and then you have a senior developer assigning tasks to agents, and you can assign them tasks on databases and maybe networks, and it comes back with different scenarios," he said.

"But you get to choose as the senior developer -- you can give them their work back to keep improving it: 'I need you to do this.' And what happens is you enable the senior developer to have more arms and legs, to extend."

Also: Most AI projects are abandoned - 5 ways to ensure your data efforts succeed

However, just because you've got AIs picking up the slack for lower-level tasks doesn't mean the role of junior developer goes away.

As my colleague Sabrina Ortiz discovered, companies need humans in the loop to ensure AIs can be trusted to work effectively. Refining human talent means allowing junior developers to learn and move into higher-level positions.

"No one becomes a senior developer just because they snap their fingers. They still need to ladder up. They must start as a junior developer. They must build up that skill. They need to understand the types of answers," said Kosla.

"Think of the apprenticeship model -- just because we have all these AI capabilities, you still need the skills to understand each layer of the stack. That's not going to go away."

4. Assess your agent's deliverables

Antony Hausdoerfer, group CIO at auto breakdown specialist The AA, said that AI agents should be tracked, traced, and assessed like their human counterparts.

"Like any member of your team, you need to gauge whether or not you can trust them, and that's generally by looking at the experience and assessing what they've done and whether or not they delivered the outcome," he said.

"You need to ask, 'Did they deliver on their commitments?' You need to start applying those assessments to the agentic world and use those as your proof points."

Also: 4 questions to ask yourself before betting on AI in your business - and why

Hausdoerfer suggested that finding the right emerging technology can be tough because the market is flooded with options. Business leaders should think carefully before trusting an agent as a member of their team.

"There are many different levels of AI now, and there's still a lot of noise in the system. There are so many applications," he said.

"You've got the business-centric systems that will improve your productivity, such as Copilot and some of the ChatGPT capabilities. But then you've got the more bespoke applications, which will be transformational from a business perspective, in terms of what AI delivers for customers."

5. Create the perfect intern

Vivek Bharadwaj, CIO at clothing manufacturer Happy Socks, suggested the rapid pace of change makes it difficult to know where to turn next when bringing AIs into your team.

"It's a brave new world," he said. "There are so many things happening. I believe that in many cases, even the definition of agentic AI is inconsistent."

Also: 4 ways to turn AI into your business advantage

To counteract this unpredictability, Bharadwaj gave his definition of agentic AI to help level-set the relationship between humans and emerging technologies in the new workplace.

"Agentic AI is, in my mind, your perfect technology intern. Rather than thinking of agentic AI as a broad superintelligence, I think of it as a personal intern you can use to automate specific parts of your workflow," he said.

"The effective person who leverages agentic AI is a systems thinker. They can understand and break down their work into discrete components, so they can say what element -- human or agentic -- contributes to the other."

Get the morning's top stories in your inbox each day with our Tech Today newsletter.