More Musings on Muse: AI Agents, Smart Data and the End of Business as Usual?
I wrote here last week about Muse, and other approaches to personal or personalised AI agents. A couple of other thoughts have come up, and I've been playing with Muse a bit more. So, time for another dive into the subject.
One conversation in particular prompted this further exploration. It was about Smart Data in home energy, and how that might pan out in terms of scheme design, schemas, consent mechanisms, trust frameworks et al.
Conventional logic, of which I have very much been part, would suggest that there are a bunch of people with relevant expertise working away to develop those frameworks, consent mechanisms, implementation roadmaps and such things. And then, at some point, a scheme would emerge, be implemented, and people and organisations would organise accordingly.
Muse and similar AI agents are going to drive a train straight through that approach. Hopefully for the better, but undoubtedly with a bit of a rollercoaster ride along the way.
Let me explain.
A decade of trying to understand my home energy data
On the back of that home energy discussion, I decided to ask Muse a few questions.
I suspected in advance that things might get interesting, not least because I'm a long-term data geek who, for many years, has been wrestling to get usable home energy data for my house.
That has included wrestling with energy providers just to get a smart meter, only to discover it wasn't particularly smart.
I've got Hive, lots of Apple HomeKit gadgetry, smart TVs, a smart car, a smart lock, and more connected devices than I care to think about.
And despite all of that, I've never had any meaningful, actionable home energy data. Not even close.
That's mainly because:
The data sources are usually not granular enough to take meaningful action.
The data certainly cannot easily be integrated across multiple sources.
The actions that might be useful are typically buried and inaccessible. Even something as straightforward as "Should I switch energy provider?" remains surprisingly difficult.
And then along comes Muse and kicks in the door of those assumptions.
Ten minutes with Muse
Here are some snapshots from my discussion with Muse.
Straight in at the deep end. Muse found my home's Energy Performance Certificate from somewhere (the property purchase, I expect). It already knew my address, and I didn't need to tell it twice.
Then things got a bit crazy.
Muse asked me to log in as ‘me’ to my British Gas online account.
I'm fine with that. All part of the Muse learning curve. And having read Jamie Smith's excellent five-part series on Muse, I knew what it was doing, why, and the implications.
I'm comfortable exploring those implications, within the context of the overarching question I raised in my last post: who does Muse actually work for?
Then... all the usage stats.
Broken down into the 30-minute windows in which they come in. Exactly the sort of granular, usable data I've been trying to access for years.
And with outliers to investigate.
Then came the killer insight.
I'll shorten the story for the sake of brevity, but Muse and I concluded that I'd accidentally left an electric radiator on in our garage twice in the previous week.
And we were right.
Now that I can see those little mistakes cost me a few pounds, I won't be doing it again.
In fact, as it's a smart radiator, I'll see if I can get it and Muse talking to each other. I certainly wouldn't rule it out...
So how about that then? Literally ten active minutes with Muse, and I've gained more actionable information about my home energy use than I had in the previous decade.
No new Smart Data scheme. No bespoke integration project. No waiting for a common user interface to emerge.
Just an AI agent, working with what was already there.
That got me thinking.
So what does this mean for Smart Data?
Here are my takes.
1. Agentic AI is going to move front and centre of Smart Data and data portability.
Especially customer-side agents.
For years, much of the focus has been on how organisations make data available, how schemes operate and how individuals navigate the resulting services.
But what happens when the individual has an agent that can navigate those services on their behalf?
That changes the starting point rather significantly.
2. Schemas and consent mechanisms: look out.
As a minimum, they need to become agent-aware.
More fundamentally, the need for people to interact directly with many of these mechanisms could reduce dramatically.
That doesn't mean the underlying standards, permissions and safeguards disappear. Quite the opposite, in some respects.
But it does mean we should be questioning whether we're designing them for a world in which humans navigate services themselves, or one in which their agents do much of that work for them.
3. The logic of multiple, separate Smart Data schemes starts to look rather different.
If there is at least one interface type at the end-user level that can navigate seamlessly across scheme boundaries, why should individuals have to care which scheme their data sits within?
Energy, banking, telecoms, mobility, health... the boundaries might matter to the organisations and regulators involved, but increasingly less to the individual.
The underlying schemes may still be necessary. But their interfaces, and the assumptions about how people use them, could change beyond recognition.
4. Trust frameworks need to put agents front and centre.
There are clearly rules required around the nature of an agent signing in to another digital service, accessing information and potentially taking action.
I wrote in some detail in my previous post on personal agents about what I think a sustainable model might look like.
Hint: it's not the "kick the door down and grab what you can see" model that Muse, and probably others, currently use.
That approach might demonstrate what's possible. It doesn't necessarily provide the foundations for a trusted, scalable system.
Which brings us back to the question of who the agent works for, what permissions it has, and what duties and obligations it owes to the individual.
An agent that can act for me is not necessarily an agent that is obliged to act in my interests.
I think that distinction is going to become critically important.
And it means that any Smart Data trust framework being designed today needs to have agents at its heart, not bolted on as an afterthought.
As for Muse and me...
I'm very aware of the potential, and probably inevitable, downsides of this approach. Not least the enshittification that could apply to any non-fiduciary personalised agent.
But I'm going to carry on pushing Muse harder, learning its limits and trying to understand what's going on under the hood.
Which is itself fascinating. Including what Muse thinks of me...
And I'm absolutely going to keep using it in the role we've seen above.
Think of it as the data portability SAS.
Breaking into the sites and apps that have been holding my data hostage, freeing it up and bringing it back home.
It won't always work. And there are some important questions about how it should work.
But we're certainly going to learn along the way.
And perhaps the biggest lesson is this: while we're busy designing the next generation of Smart Data schemes, AI agents are already demonstrating what the next generation of Smart Data experiences might look like.
We should probably be paying attention.
Want to try Privacy Pulse yourself for Free, scan your own website and see what score and grade you get?