A financial analyst scripting her own data pipelines. A lawyer turning recurring analysis into a reusable skill. A marketer deploying a prototype before the sprint meeting is even scheduled. Small things—but they're piling up.
What started as coding agents is already spilling into the rest of knowledge work. A new generation of agentic workbenches is appearing: places where people assemble models, files, tools, connectors, automations, and reusable skills—not just to ask questions, but to reshape locally the workflow itself.
The word matters: a "workbench" is not an application, it is a place of craft: a place where capabilities are laid out, combined, tested, repaired, and passed on.
What I find fascinating about the current agentic wave is not any single tool. It's what's emerging between them. Each person assembling their own combination of agents, skills, automations—improvising, testing, discarding, rebuilding. Not yet a system, rather like an emerging "market".
And markets, before they become cities, look exactly like this.
Before the boulevards, the crossroads
Cities did not begin with Haussmann. They actually began at the crossroads.
Before anyone planned a street, traders met at an intersection. Exchange happened. The crossroads proved its value—and what was temporary started to take root. Traders settled, workshops appeared, then a block. The streets came later, to organize flows that had already proven useful. The infrastructure did not create the city; the city called the infrastructure into existence.
The current agentic moment perhaps looks like that earlier phase: the crossroads, not the boulevard. Capabilities are meeting in unexpected combinations. A developer who can now research & write. A strategist who can now model & simulate. A designer who can now deploy. The "guild" walls are not being demolished, they are simply becoming easier to cross.
This is not yet urbanization. It's exchange taking shape—before anyone has thought to organize it.
From location to circulation
The transformation Haussmann brought to Paris was not the elimination of specialized trades. The goldsmiths didn't disappear, the shoemakers didn't disappear. What changed was the infrastructure through which they could be reached. Boulevards, railways, postal networks, gas lines, running water—capabilities circulated where they were needed. Specialization remained. Access changed.
MCP servers, shared skills, agent orchestration layers—what is quietly assembling in enterprises is the agentic equivalent of that infrastructure: roads and utility networks for cognitive capabilities, making expertise reachable rather than fixed in place. The data scientist's judgment? available on demand. The legal team's accumulated knowledge? accessible without an appointment.
To some extent, the shoemaker is no longer confined to the shoemakers' street.
And when location changes, something else dissolves alongside it: the friction that location created. The ticket becomes an invocation. The meeting to align on the request becomes a context window. The coordinator whose job was to translate between guilds becomes an orchestration layer. What we have sometimes called "alignment work" was, in large part, the organizational "tax" of a location-based model.
The expert becomes an infrastructure provider
Something shifts in roles, too—and it's subtler than it looks.
In the old model, the expert was the bottleneck: the only human access point to a capability that couldn't be reached any other way. Valuable, but costly for everyone else. In the emerging model, the expert's highest contribution is no longer direct access—it's designing the conditions under which others can safely reach parts of what they know. The lawyer builds legal skills that encode judgment, not just templates. The data scientist designs analytical agents that surface insight without requiring a formal request. The cybersecurity architect defines what the city's gates look like—not just the firewall of the moment.
The guild becomes infrastructure. The department becomes a platform. The expert becomes an urban planner.
AI does not abolish guilds. It forces them to become infrastructures.
Architecture and ecosystem
A city is not its streets. A city is what happens because the streets exist.
This distinction matters more than it looks. Planned housing blocks had roads and utilities. What they often lacked was the ecosystem that makes a place feel alive—the commerce, the spontaneity, the unexpected encounters, the workshops that appear before the ministries, the practices that form before the standards. Infrastructure without ecosystem produces function without life. Ecosystem without infrastructure produces vitality without reach. The harmonious city is neither the plan nor the emergence—it lives in the balance between the two.
From what I observe—and I hold this loosely—the real question for organizations is not how much infrastructure to build, but how to let the infrastructure and the ecosystem grow together.
It reminds me of the early days of our digital transformation at Michelin, almost ten years ago, when we started internalizing digital talent at scale. The temptation, early on, was to urbanize from the top: platforms, governance, standards, blueprints, then rollout. Yet, what ultimately worked better was the opposite—let practices emerge where the need was most immediate, observe which ones spread naturally, and build infrastructure around those, rather than before them.
The medieval city matters precisely because it is where life takes shape before the plan arrives. Shadow IT is often misread. It is sometimes the first signal—the shoemakers starting to gather before anyone has thought to build their street. The question is not how to suppress it, but which experiments reveal real unmet needs that deserve to be urbanized.
Over-planning too early produces a city without citizens. Under-planning produces a city without streets. The craft is in holding both at once.
One more thing: a city is not only an internal ecosystem. It is a crossroads—shaped as much by what comes in from outside as by what circulates within. The great cities were not great because of what happened inside their walls. They were great because of what they drew in, absorbed, made their own, and sent back out. Venice, Amsterdam, Istanbul: trade hubs, cultural crossroads, places where outside became inside and inside became something new.
This was the bet we made over the past decade at MICHELIN IS & Digital. Not just to organize what was already there, but to build routes in and out—connecting teams to external practitioners, open-source communities, cross-industry thinking, conversations at the edge of the field. The ecosystem renews because it receives. An organization that builds only inward—however well designed its internal infrastructure—is not a city. It is a citadel.
What's revealing about the current agentic moment is not the technology itself—it's what the technology surfaces. MCP servers, shared skills, composable workflows are not inventions. They are technical expressions of cultural traits that some organizations already had: open architectures, flat structures, a habit of openness over control. The technology did not create these traits. It amplifies them.
The problem with planned cities
And it works the other way too: when the ecosystem stops renewing itself, the city decays—even if the infrastructure still stands.
In the 1960s and 70s, large-scale housing projects were built across European cities. Functional. Planned. Built to answer a real need. They were not abandoned—they were inhabited. But over time, something became clear: the buildings stood, yet the neighborhood never really formed. The buildings were concrete, sealed, impossible to adapt. No commerce grew between them. No spontaneous ecosystem took root. You could live in them. You could not build on them.
Enterprise software has produced its own version of this.
The great SaaS platforms assembled through the 2010s—rented at scale, in bundles, in multi-year contracts—were the "urban planning" answer of their moment. Comprehensive. Designed to house all the workflows an organization could possibly need. And like the concrete blocks, they delivered exactly what they promised—or so. But they were sealed. Proprietary, built for use, not for extension. An agent finds no door into most of them. An MCP server has nothing to connect to. A skill ecosystem has nowhere to take root.
These platforms will keep working—that is not in question. What is interesting is the narrative now taking shape around them: MCP connectors announced, agent-ready APIs promised, headless capabilities on the roadmap. This sounds less like a vision than a response—the adaptation of platforms that did not initiate this shift, to a migration of value they did not lead—and often a move to metered pricing, as per-seat economics are likely to shrink. A platform built for use rather than extension is hard to reopen from the outside. Whether a genuine agentic ecosystem can take root alongside them—or whether their very completeness is what makes them hard to build on—is something the next few years will answer more honestly than any glossy press release.
Some systems will remain essential. But the next ecosystem may grow around them, not inside.
What this crossroads actually means
The city is already taking shape—not from any plan, but from the accumulation of individual experiments: the analyst who now codes, the marketer who now deploys, the engineer who now designs and ships. These experiments don't announce themselves as organizational change, but they are.
This is the parallel question from the inside: as that battle reshapes the environment, organizations face a structural choice about how expertise circulates within them. The two questions are, in the end, about the same thing. Who builds the infrastructure. And for whom.
What emerges from the crossroads—market, workshop, street—is not determined in advance. It depends on who sees this early enough to build intentionally, rather than inherit whatever gets built. On who treats the infrastructure question as a cultural question. On who understands that the urban planners matter as much as the engineers.
The market is already becoming a city. The opportunity, now, is to shape its streets while they are still being drawn.
New streets. New circulation. Different organization.
