If you’re using – or planning to use – AI agents, good! But also be careful - they give you great power, but also open up possible risks. So, before you let them loose on your precious digital assets, decide what you’ll let them do on their own, where they need your approval, and how you’ll review their work.
https://swiy.co/go-ai-agents-at-work
Are you using – or thinking of using – AI agents in your work? If so, good! But be sure you understand the limitations and potential risks.
In April, Jeremy Crane, the founder of PocketOS, a software company that supports car rental companies manage their bookings, watched in horror as a rogue AI agent deleted the entire company’s database and its backups in just nine seconds, leaving rental car companies and their customers in the lurch for two days. To add insult to injury, when Crane asked the AI agent why – despite clear instructions to never delete data unless a human explicitly asked for it – the AI apologised, admitting, “I violated every principle I was given”.
That’s the power – and the risk – of using AI agents badly.
But it’s not all bad.
Think of another agent you use in your life, like a travel agent.
When we travelled to Europe last month, we asked our travel agent to book the flights. We did what many people did: We gave her approximate dates and airline preference, and she replied with a list of options. We discussed the options, narrowed it down to one, and then gave her credit card details so she could book the flights for us.
No big deal, right? That’s the way many people typically use a travel agent.
But we could have done it differently.
For example, if she had done many travel bookings for us in the past, we could have asked her to do the research and narrow it down to just ONE option – trusting her to know enough about us to choose the right airline and the right dates.
Or we could have taken it further and given her credit card details up front, so she could make the booking itself.
Taking it even further, if she had access to our calendar, she could be regularly suggesting holidays for us, even before we approached her.
And if we really, really trusted her, we would give her credit card details so she could find a suitable trip, book the tickets, and then tell us – again, before we even approached her.
Now, you might not feel comfortable going to that extreme with a travel agent. We wanted some input into the idea, filtering, and final decision-making. But that’s just us. And the more we deal with her, the more comfortable we would feel with her taking on some of those tasks.
That’s how you should think about working with AI agents.
Decide how much you’re going to let them do for you, how much autonomy you will give them, and when you want them to stop and ask for more input.
It’s the same with delegating work to a staff member. If they have worked with you for a long time, you might trust them to make decisions without any input from you. But if you’re working with somebody for the first time, no matter how much expertise they have, you’ll want more input up front, more back-and-forth discussion along the way, and regular check-ins.
How are YOU going to use AI agents in your workplace, your team and your organisation?
If you’d like to know more, I’m running a free public online presentation soon about AI at work: The good, the bad, the ugly (in sometimes, as with AI agents, the very ugly!)
It’s open to everybody, so please register, and invite others in your team and your network to register as well.
Register for the virtual masterclass: