In a recent post, we painted Subset as providers of custom information couriers, entities that distribute saved things to high relevance targets in a trustworthy, autonomous, performant way. We then examined an inherently faulty assumption:
In the physical world, couriers are confident that their environment and its systems are not impeding or acting maliciously against them
In the platform world, that doesn't hold. Platforms penalise user and content migration and take steps to shut down the promiscuous, protocolised distribution of interesting things. This creates an impasse.
Platforms want to "protect" end users, monopolise their attention and extract most of the value they generate in the process. This is a historically profitable activity, so platform providers are ready and willing to distort reality to preserve it. End users, in contrast, are beginning to chafe at the excessive friction, exit restrictions and invasive access they're compelled to endure digitally as they pursue their analog ends (talking to people, maintaining relationships). Fortunately, a force is breaking the stalemate: Unix-like, Boydian burglar agents.
The Unix philosophy "emphasises building simple, compact, clear, modular, and extensible code that can be easily maintained and repurposed by developers other than its creators." This can be contrasted with the "everything" philosophy—examples include Amazon as the "everything store", Notion as the "all-in-one workspace", and Apple's quest to create the operating system for their user's lives. A Unix-like agent is highly competent in a limited area. An everything agent is highly competent in all areas. Unix-like agents already exist. Everything agents do not and may or may not be arriving soon.
In part, this non-existence of everything agents is a capability issue. The systems that carry the operating load of contemporary society are socio-technical in nature. If those operations could be reduced to mere functional requirements, everything agents would flourish. They can't (outside of contrived scenarios) so they won't (for now).
The other issue facing everything agents is privacy and security. Doing everything on behalf of a user implies user-level access to everything. This is a problem—for a case study see Microsoft's Recall climbdown. Unix-like agents, however, escape the capability and access issues. They're designed to be constrained and to conduct limited, measurable tasks in an auditable way.
Current conceptions of Unix-like agents include logging and monitoring solutions, and the 2nd party partnerships and 3rd party integrations permitted by platform overlords. However, the next generation of Unix-like agents is near. They'll have a little more scope, afforded by increasingly protocolised operations. And they will be unyoked from centralised providers making a "I solemnly swear that I'm up to no good" privacy and security vow that fails because of incentive design and "market conditions".
With the transition to a protocolised world and the proliferation of multi-modal models, agent capabilities are expanding to incorporate decision making and intent, instead of pre-defined deterministic outcomes alone. This capacity to exhibit intent and make a decision requires some mechanism for carving through uncertainty and making the best choice with limited information.
In platform world, such decisions could be traversed via instruction. The explicit, upfront enumeration of scenarios and responses. If A, then B. If C, then D. In the protocolised world, such decisions are made in a Boydian fashion. A mission is defined—get this thing to that person. Underlying the mission is user-agent trust—the agent is uniquely aspected to the user themselves— and an existing context—some things about that person are known. That is almost enough to accomplish the mission.
This is where we come to the burglar mindset. In Nakatomi Space, Geoff Manaugh writes about Die Hard and the inappropriate use of the built environment:
"What I find so interesting about Die Hard—in addition to unironically enjoying the film—is that it cinematically depicts what it means to bend space to your own particular navigational needs. This mutational exploration of architecture even supplies the building’s narrative premise: the terrorists are there for no other reason than to drill through and rob the Nakatomi Corporation’s electromagnetically sealed vault.
Die Hard asks naive but powerful questions: If you have to get from A to B—that is, from the 31st floor to the lobby, or from the 26th floor to the roof—why not blast, carve, shoot, lockpick, and climb your way there, hitchhiking rides atop elevator cars and meandering through the labyrinthine, previously unexposed back-corridors of the built environment?
Why not personally infest the spaces around you?"
This went on to the become A Burglar's Guide to the City:
"For the past several years, I’ve been writing a book about the relationship between burglary and architecture. Burglary, as it happens, requires architecture: it is a spatial crime. Without buildings, burglary, in its current legal form, could not exist. Committing it requires an inside and an outside; it’s impossible without boundaries, thresholds, windows, and walls. In fact, one needn’t steal anything at all to be a burglar. In a sense, as a crime, it is part of the built environment; the design of any structure always implies a way to break into it."
In the platform world, law-abiding citizens mostly remain within their preferred platform's walls. Their friends are there and it's easy to communicate. But when they want to go cross-platform, however, there's a fuss. There's walls, gates, checkpoints, carry restrictions, tracking. Enough impedance to squash recurrence of migration activities and enforce a platform sedentariness.
In the protocol world, however, citizens are platform-agnostic and have burglar agents acting as extensions of their intentions. "Here is a thing; get it to my friend; they're somewhere, somehow, online." The courier then engages McClane mode in order to make the delivery:
"McClane explores the tower—called Nakatomi Plaza—via elevator shafts and air ducts, crashing through windows from the outside-in and shooting open the locks of rooftop doorways. If there is not a corridor, he makes one; if there is not an opening, there will be soon.
Over the course of the film, McClane blows up whole sections of the building; he stops elevators between floors; and he otherwise explores the internal spaces of Nakatomi Plaza in acts of virtuoso navigation that were neither imagined nor physically planned for by the architects.
His is an infrastructure of nearly uninhibited movement within the material structure of the building."
To summarise: UBB agents—Unix-like, Boydian burglar entities—possess:
A close alignment to a singular user or entity and their actions
A deliberately constrained set of capabilities and protocol-savvy
A mission-driven approach to objectives
An ability to imaginatively manipulate the space they traverse
They're the entities that will break the deadlock between digital platforms, the rise of protocols, and the relationships that end users have always prioritised over any particular digital means. They're the things that will free up the flow of Dunbar goods, empower curators and expose many more people to great search outcomes.
This is a big part of what we're working on at Subset; trustworthy, performant and autonomous agents that can accomplish saving, sharing and search missions on behalf of the users and entities they're uniquely aligned to. Because the question is not if UBB agents will arrive, but when. Our bet: sooner rather than later.
