A few years ago I was in Altyn-Arashan, an alpine valley a few hours from Karakol in Kyrgyzstan, soaking in a thermal bath after a long trek. The bathhouse is nothing. Wood planks, a concrete floor, a rusted pipe running straight from a natural spring into the pool. Forty to fifty degrees, rich in minerals, no design, no branding, no interface. Somebody just ran a pipe from where the hot water is to where people needed it.
Nobody photographs the pipe. Everybody depends on it.
I keep thinking about that pipe because of something that happened over eight years at Salesforce.
I spent eight years at Salesforce. Some of the best years of my career.
One of the things I’m most proud of is the work we did on the automation platform. Flow Builder, process automation, event-driven architecture, IoT Cloud. At the time it felt like plumbing. Important plumbing, but plumbing. Connecting data to action. Letting a business say “when this happens, do that” without writing code.
What I understand now is that we were building the foundation that makes Agentforce possible. And the same foundation is what made Claudeforce possible too. The partnership announced a few weeks ago brings Claude’s reasoning together with Salesforce’s data, workflows, business logic, and governance so an agent can take real action wherever people work. That’s not a coincidence. An AI agent can sit inside Salesforce and do something trustworthy because of the years of work that went into the layer underneath. The data model. The permissions engine. The workflow logic. Without that, Claude has nothing solid to work with.
I’m proud of that work. It aged well. Big kudos to the whole team, and to those of you still building the platform engine at Salesforce or Mulesoft.
The other thing I worked on at Salesforce was Lightning. The complete redesign of the UI and UX platform. New design system, new component model, new way of building applications on top of Salesforce. Every app in the ecosystem rebuilt around it. It was a massive effort, and it defined how people interacted with Salesforce for years.
It might continue to. But I find myself wondering for how much longer.
What the Interface Is Actually For
When Marc Benioff introduced Claudeforce, he said “the UI is the AI.” It’s a big thing for a company to say about its own product. It’s an admission of where the interaction layer is going.
When an agent can sit between a user and a system of record, find the right information, take the right action, and close the loop without the user ever opening an application, what is the designed UI for?
My answer: governance, configuration, and oversight. The screens become less where work happens and more where work is set up, reviewed, and controlled. That’s still valuable. But it’s a different kind of value than what enterprise software has been selling for the last twenty years.
The automation layer we built becomes more valuable in this world, not less. It’s what the agent depends on. The experience layer, the navigation, the designed screens, the way humans used to move through software to get things done, that is the part under pressure.
This Is Bigger Than One Company
Salesforce is not the only company facing this. I’ve watched the same dynamic at other large enterprise software organizations. ERP platforms, HR suites, finance tools, CRM systems built over decades. All of them share the same assumption: a human sits at a screen and moves through a designed experience to do their work.
That assumption is weakening. GPT-6 Astra, released ten days ago, is built not just to reason but to operate. It browses, it uses apps, it fills forms, it updates records, it works inside spreadsheets and document editors and professional software. Previous models told you how to do things. Astra is built to do them.
This is the direction. Not one model, not one partnership. Every major lab is moving toward agents that don’t advise the person at the keyboard but do the work directly. The interface in between becomes less necessary with every release.
The companies that understand this are asking the right questions. What do we own that an agent needs? Where is our data irreplaceable? What compliance and governance role do we play that a new entrant would take years to rebuild? What stays valuable when the UI and more of the workflow move somewhere else?
The ones leaning into that layer, the data, the permissions, the integrations, the trust built over years, have a real path forward. The ones still betting mainly on the interface are in a harder position. Not immediately, not catastrophically. But the direction is clear and the pace is picking up.
And we are going to see more deals like Claudeforce. The model labs need the data and the trust infrastructure that enterprise software companies spent decades building. The enterprise companies need the intelligence layer to stay relevant. Neither side can build what the other has quickly. More partnerships are coming, and each one will move a little more of the work out of the application and into the agent.
The Part That Doesn’t Change
The tools are moving fast. Agents are getting dramatically more capable. Astra can run multi-step professional workflows, produce finished documents, check its own work, and operate software end to end. That’s real.
But none of these tools know what to build or why. They don’t know which workflow matters to your customer and which one doesn’t. They don’t know what your compliance obligation means in practice, or which integration is quietly holding everything else together. That judgment is still human. And it’s getting more important, not less, because the implementation is getting easier and the decisions are getting more consequential.
The engineers and product people who do well from here are the ones who understand systems, not just codebases. Who decide what to build next instead of only executing what they’re told. Who stay engaged with what the AI is doing instead of accepting its output.
Two Platforms, One Lesson
I helped build two platforms at Salesforce. One was the plumbing nobody saw. The other was the interface everybody saw. I was proud of both, and at the time I would have told you the interface was the harder, more visible, more important achievement.
Eight years later, the plumbing is what agents run on. The interface is what agents are starting to replace.
The layer where work happens is moving. The foundation underneath is what lasts.
Based on these learnings, and while exploring and adapting to this new way of working and creating products, I’m building ARTEX, a living digital art platform, from Milan.
More at codame.com