I’ve always been an avid watcher of how people use technology out in the wild, the devices, the form factors, the increasing lack of headphones and self-awareness. On a recent journey into London, I was taken by how far we’d actually come in the last decade or so with many commuters, now using a single device (probably a phone rather than a laptop) for the entire journey. And then it occurred to me how much worse it used to be. Somewhere around 2010, I remember every table on that same route having the same kit: four devices, and a bulging rucksack with four charging cables and a small nest of adapters to keep them all alive. Two phones, one they’d chosen and clearly coveted, kept in a padded sleeve and brought out with much reverence (“my precious...”), the other issued by the company and chucked in the bottom of the bag. Two computers, a sleek tablet they actually used and a boxy slab of beige they carried because they had to.


I remember sitting there thinking, where did it all go wrong? What happened next is so familiar we’ve forgotten it was ever a fight. People started bringing their own. They had better technology at home than they were given at work, and anyone with an iPad on the sofa and a grey box on their desk knew precisely which was the future and which was a Victorian filing cabinet with a screen. “Why does our procurement system look like it’s running on a mainframe,” they asked, “when I can shop at Amazon with a couple of taps?” It was also cheaper for the business to let them. The objection was never whether it was a good idea, only how to let an unknown device onto the network and keep things secure. For a while the honest answer was that we couldn’t. Then device management and policy arrived, and the argument simply ended.


I wrote about what came next in Business Reimagined, back in 2013. The consumerisation of IT, I argued, had a second wave coming, and it wouldn’t be about devices but services: the tools people used in their personal lives and, more importantly, the principles underneath them. It started with people wanting the open, share-first way those services work inside the business, years before Slack and Teams made that normal. At home our default is to be open and keep a few things private. In business it has nearly always been the other way round, because knowledge is power for whoever holds it. We happily shared our personal lives on social media, while in our organisations, useful information lay locked away in inboxes, unreadable by others by design. I also said that if you can’t trust employees with these tools, you can hardly trust them without them. That was generation two. The first brought a device and the second brought a service.


In May, in The Elephant Path, I told the story of the audit partner who bought his own wireless kit at lunchtime (the classic elephant path, the scruffy track worn across the grass by people who found a faster way to where they needed to go), and said Shadow AI was the Wi-Fi moment all over again. I think it’s bigger than that. This is the third time we’ve seen this movie, and what’s walking through the door is different.


The third wave of the consumerisation of IT isn’t a device or a service. It’s an agent, and potentially a whole army of them. I have about ten in daily use, spread across multiple platforms (OpenAI, Anthropic, Microsoft and Google, plus some I self-host), and I wouldn’t want to work without them. My main one, the ubiquitous “chief of staff”, has replaced email as the first thing I check in the morning. It holds months of context, preferences and corrections that nobody can buy off the shelf. It’s much closer to a colleague than a tool, so the question is no longer “can I bring my phone?” It’s “can I bring my colleague?”


Deloitte asked pretty much that question in October, citing its own research that 49% of workers using AI at work already use tools their employer hasn’t approved.


It’s the same story in one way: your personal use of AI at home is likely better than what you’re given at work, because nobody has to approve it, you’re not constrained by data governance, and it rewards time spent on personalisation. Your agent knows you in a way your company never will. But it’s different in two ways that matter a great deal.


An agent acts. A device or service held your data. An agent reads your inbox, books your travel and replies to your supplier, and to the company’s systems it looks exactly like you. The first wave asked what was on the device, the second what was stored in the service. This one asks who is acting, and on whose authority. Identity management helped us solve the first wave, and we can learn from that. What’s new is that this isn’t the device or the person, but something acting on the person’s behalf.


It knows the whole of you. We fixed the first wave by separating work from personal on the same device, and very little was lost, because the value of a phone was having one device, not having your work and personal data talk to each other. An agent is different. The value lies in the two talking to each other. Mine knows my diary, my job, my family, my tax return and my fridge, and that is the point of it. The corporate alternative will only ever know the work half of me. Partition it in the name of security and you’ve rebuilt the corporate agent. And an audit trail on my agent is like inviting my boss into my house before I’ve had time to dust.


And each wave has been harder to see than the last. You could always spot a bring-your-own device: new, smear free and worshipped, on the table in front of you. The services were harder to spot (the WhatsApp group that exists because it’s easier than SharePoint because you didn’t need a letter from your Mum to give you permission to set up a workspace) but you could find them if you went looking. An agent doesn’t sit anywhere. Something you held, then something you used, now something that acts for you. As I said, I have about ten. How many do you think are quietly sitting behind your colleagues? I’ll bet it’s not none.


So will organisations hold the line? Some will, and for some kinds of data they absolutely should. But a blanket ban is another thing altogether. To the machine, it says you’re a tool, and tools are issued, not chosen. Worse, to the person, it says we don’t trust you to use your own judgement. That’s our old friend the robot frame, applied twice - low expectations of people produce exactly the limited engagement those expectations demand.


There’s a talent question hiding in here too. Personally, I would (and do) think twice about working for any organisation that was unnecessarily (and I stress unnecessarily) restrictive about how I use AI. On a good day, I’d be asking myself why I was being asked to leave the thing I’ve spent months shaping to make me better at my job outside the door. On a bad day, I’d be thinking I simply don’t want to work somewhere that doesn’t want to have me at my best. The people using their own device didn’t wait for permission, and the people arriving with their own AI now won’t either. And I don’t mean people who are quick with a prompt. I mean the ones who form their own answer first and then let the agent challenge it. We’d be lucky to find them. Ask someone like that to leave their colleague at the door and they may well just leave too.


What I’d suggest instead is the answer that worked last time: build the architecture, and think about it the way we already do for people. When a talented new colleague joins, you don’t hand them the keys to everything on day one. They get an identity and clear lines of authority, and they’re held accountable for how they and their agents behave.


  • Identity: the agent needs its own, related to mine but distinct, so you can tell what I did from what it did.
  • Authority: what it may read, do and spend, set by the organisation and widened as trust is earned, the way the banks built guide rails for card-not-present payments.
  • Accountability: I can delegate the task, but I cannot and should never delegate the accountability. If it books the wrong flight, that’s my name on it.


And the audit trail? Audit the door, not the house. The organisation has every right to see what my agent did to its systems, and no business inside the rest of the house.

Three waves of consumerisation in, and the movies are more or less the same. The people doing the work get there first. The organisation then decides whether to build the road or pretend the grass isn’t trampled.


What’s different this time is what’s walking across it.


I often refer back to a formula: effective transformation equals employee engagement multiplied by transformational technology, a product, not a sum. A person who arrives on day one with an agent they have shaped over a year is bringing both: the engagement and the technology, together. An organisation can multiply that, or it can ask them to hand it in at the door.


It isn’t a technology decision. It’s a decision about what you expect of the people you’ve hired, and of what they’ve learned to work with. What will you choose?