I’ve been popping up in a few places lately. Here’s a round-up: a talk, an academic paper, and a blog post.
Rethink AI, WIRED x Kendryl
I spoke about AI agents as part of a Wired event called Rethink AI with Azeem Azhar and others (as previously mentioned).
Here’s the Rethink AI homepage where you can find all the vids.
It’s sponsored content (thanks Kendryl) but that’s no bad thing, it means I got make-up, used proper teleprompter for the first time (with someone driving it!), and the set was sooper nice.
As a talk focused on future business impact and what to do today I wanted to help non-technical folks understand the technology through examples, extrapolate to where it’s going, and give practical C-suite-level pointers on how to prepare, in three areas:
- Your customers are moving to chat and your business risks becoming invisible
- Agents will be everywhere, intelligence is a commodity, and what matters is access to systems
- Self-driving corporations are the destination… and you can start experimenting today.
(I used self-driving vending machines as an example, and just a few days later Anthropic came out with theirs! Hence my recent post about business takeaways from autonomous vending.)
Watch the talk on YouTube: AI Agents: Your Next Employee, or Your Next Boss.
Please do share it round.
Star-Painters and Shape-Makers
I inspired a chapter in a book!
The backstory is that back in October 2023 I started putting single-purpose AI cursors on a multiplayer online whiteboard.
I keep coming back to these experiments. The reason is that identity helps us attach capability and knowledge to bundles of functionality, a necessary antidote to the singular ChatGPT that is utterly obscure about what it remembers and what it can do.
We don’t need to anthropomorphise our AI interfaces – we can get away with way, way less. I call that minimum viable identity (Feb 2025, see the bottom of the post).
ANYWAY.
I was playing with these ideas when I met Professor Jaime Banks (Syracuse University). I gave her a corridor demo, we talked some.
It obviously made an impression because that demo became the opening of an insightful and so generative short chapter in Oxford Intersections: AI in Society (edited by Philipp Hacker, March 2025).
You’ll only be able to read it if you have access via your institution, but here’s the link:
Star-Painters and Shape-Makers: Considering Personal Identity in Relation to Social Artificial Intelligence, Jaime Banks (2025).
I have the full article, and I’ll give you the first couple of paragraphs here by way an of intro…
On the backdrop of a hustling, buzzy conference in early 2024, serendipity found my path crossing that of Matt Webb–maker, thinker, and engager of “weird new things.” Matt was demonstrating PartyKit, an open-source platform for apps including some supporting artificially intelligent agents. This demo comprised a split screen–on one side a whiteboard drawing app and on the other a chat interface housing a number of what he calls NPCs (or non-player characters, from gaming parlance) that may or may not be driven by formal AI. In a collaboration between user and NPCs, activities unfold in the draw space-and each NPC has a specific function. One might be designated for painting stars, another for creating shapes, and another for writing poems or making writing suggestions. Based on these functions, an NPC could be recruited to help with the drawing, or it could autonomously volunteer its services when a set of conditions manifests (e.g., when the user draws a star, the star-painting NPC says, “I can paint that!” See Webb [2023] for a narrated demo).
What I recall best from that day was my reaction to the demo–and then my reaction to my reaction. I was seeing each of these NPCs-inferred entities represented by circles and text in the chat and actions in the draw space. Each had something that made it seem qualitatively different from the others, and on contemplation I realized that something was each entity’s function, how the function was expressed, and all the things I associate with those functions and expressions. I saw the star-painter as bubbly and expressive, the shape-maker as industrious and careful, and the poet as heady and dramatic. It struck me how remarkably simple it had been for the NPCs to prompt my interpretation of them as having effective identities in relation to one another, parceled out by functions and iconic avatars. My fascination wandered: What is the minimum viable cue by which an AI might be seen as having a personal identity–a cue that differentiates it from other instances of the same effective form of AI? What are the implications of this differentiation in human-machine teaming and socializing scenarios? What might these identity inferences mean for how we see AIs as being worth recognition as unique entities–and is that recognition likely a self-similar one or a sort of othering?
The first section is called What Is Identity Anyway? and from that point it gets really good. I will be mining that text and those references for a long time to come.
I want to quote one more time, the closing lines:
Once an AI has access to sensors, is mobile, and must plan and evaluate its own behaviors, that begins to look like the conditions required for an independent and discrete existence–and for the discrimination of self and other. The star-painter may know itself apart from the shape-maker.
/swoons
This is always what I hope for with my work – that it might, even in a small way, help someone just a step or two on their own journey, and perhaps even spark a new perspective.
There are so many great jumping off point in Banks’ chapter. Do check it out if you are able and you can find all the references listed here.
Tobias says something kind
I got a mention in Tobias Revell’s latest blog post, Box131: You’re a National Security Project, Harry.
He talks about my unpacking of multiplayer AI and conversational turn-taking and then says:
The solutions Matt walks through are elegant in that vibey/London/Blue Peter way that he’s great at – none of that Californian glamour, just gluesticks and tape but goddamnit it works and has potential to work.
And this is again something I aspire to with all my work, and thank you so much Tobias for saying!
(And then he takes the ideas somewhere new which makes me think something new - prompt completions as sensory apparatus - and that might be the seed of a future thing!)
There is so much tall talk around technology and it’s deliberate because it creates this priesthood, right; it creates a glamour of value and also dissuades questioning.
But you can always break down something that works into unmagic Lego bricks that anyone can grasp and reason with. And I love doing that, especially when I think I’ve hit on something which is novel and could lead somewhere.
Will be adding that one to my brag list.
I’ve been popping up in a few places lately. Here’s a round-up: a talk, an academic paper, and a blog post.
Rethink AI, WIRED x Kendryl
I spoke about AI agents as part of a Wired event called Rethink AI with Azeem Azhar and others (as previously mentioned).
Here’s the Rethink AI homepage where you can find all the vids.
It’s sponsored content (thanks Kendryl) but that’s no bad thing, it means I got make-up, used proper teleprompter for the first time (with someone driving it!), and the set was sooper nice.
As a talk focused on future business impact and what to do today I wanted to help non-technical folks understand the technology through examples, extrapolate to where it’s going, and give practical C-suite-level pointers on how to prepare, in three areas:
(I used self-driving vending machines as an example, and just a few days later Anthropic came out with theirs! Hence my recent post about business takeaways from autonomous vending.)
Watch the talk on YouTube: AI Agents: Your Next Employee, or Your Next Boss.
Please do share it round.
Star-Painters and Shape-Makers
I inspired a chapter in a book!
The backstory is that back in October 2023 I started putting single-purpose AI cursors on a multiplayer online whiteboard.
I keep coming back to these experiments. The reason is that identity helps us attach capability and knowledge to bundles of functionality, a necessary antidote to the singular ChatGPT that is utterly obscure about what it remembers and what it can do.
We don’t need to anthropomorphise our AI interfaces – we can get away with way, way less. I call that minimum viable identity (Feb 2025, see the bottom of the post).
ANYWAY.
I was playing with these ideas when I met Professor Jaime Banks (Syracuse University). I gave her a corridor demo, we talked some.
It obviously made an impression because that demo became the opening of an insightful and so generative short chapter in Oxford Intersections: AI in Society (edited by Philipp Hacker, March 2025).
You’ll only be able to read it if you have access via your institution, but here’s the link:
Star-Painters and Shape-Makers: Considering Personal Identity in Relation to Social Artificial Intelligence, Jaime Banks (2025).
I have the full article, and I’ll give you the first couple of paragraphs here by way an of intro…
The first section is called What Is Identity Anyway? and from that point it gets really good. I will be mining that text and those references for a long time to come.
I want to quote one more time, the closing lines:
/swoons
This is always what I hope for with my work – that it might, even in a small way, help someone just a step or two on their own journey, and perhaps even spark a new perspective.
There are so many great jumping off point in Banks’ chapter. Do check it out if you are able and you can find all the references listed here.
Tobias says something kind
I got a mention in Tobias Revell’s latest blog post, Box131: You’re a National Security Project, Harry.
He talks about my unpacking of multiplayer AI and conversational turn-taking and then says:
And this is again something I aspire to with all my work, and thank you so much Tobias for saying!
(And then he takes the ideas somewhere new which makes me think something new - prompt completions as sensory apparatus - and that might be the seed of a future thing!)
There is so much tall talk around technology and it’s deliberate because it creates this priesthood, right; it creates a glamour of value and also dissuades questioning.
But you can always break down something that works into unmagic Lego bricks that anyone can grasp and reason with. And I love doing that, especially when I think I’ve hit on something which is novel and could lead somewhere.
Will be adding that one to my brag list.