After sitting through the keynote at Salesforce’s developer conference, TDX, in San Francisco I was a little shell-shocked. There were lots of announcements, several detailed demos, and a whole litany of product names and technical features I tried – and failed – to hold in my head.
But after some time to reflect I realized that none of it mattered.
Because what Salesforce really announced at TDX was not a list of product updates, but just how serious it is about transforming for the agentic era.
The backbone of the keynote was ostensibly a new platform called ‘Headless 360’ – an infrastructure that makes Salesforce’s portfolio of data sources, actions, and tools available via APIs, MCP servers, and skills. All without any hint of a user interface.
As Joe Inzerillo, President Enterprise & AI Technology, put it:
Every one of these [Salesforce] systems is represented in these endpoints. And these endpoints are what you're going to use – or more accurately, agents working on your behalf are going to use.
On the surface, that reads as a technical move. A more flexible way to integrate existing functionality into new systems – and agents – so that you can extend and build on the platform.
But in my opinion, that’s not really the story at all – not by a long way.
Instead I view this refactoring as evidence that Salesforce is serious about the shift it is evangelizing towards the agentic enterprise – and that it is basically showing its customers what that shift looks like in very real – and very concrete – terms.
Not because it is adding agents, but because it is restructuring itself to be legible to them.
And Headless 360 is the mechanism for doing that.
Can’t get you out of my head
Historically, Salesforce functionality has been consumed via applications. Users log in, navigate screens and follow predefined flows. And so it has been forever, pretty much.
Even as Apex, AppExchange and Lightning – for example – extended the scope of the platform over time, the center of gravity remained the same. Salesforce was a place where people went to do their work.
But headless removes that assumption by exposing the capabilities that sit behind those applications directly – which means data can be retrieved, actions invoked and processes triggered without going through the interface. Flipping it from an integrated work surface to a set of tools through which work can be facilitated.
By exposing its capabilities via Headless 360, Salesforce is making them legible to agents, effectively reprogramming its operating model rather than its products. It is making its fundamental capabilities available in ways that don’t require a human.
It is, bluntly, re-optimizing its platform for agents, not people.
Because agents don’t need applications and they don’t move through screens or follow prescribed flows. They consume context, identify what needs to be done, and then select capabilities as tools to act. Which only works if those capabilities are exposed in a form they can actually use.
Seen through that lens, the rest of the keynote becomes easier to interpret.
Heads will roll
Vibe coding, agent builders, Claude Code integration, testing frameworks, deterministic harnesses and marketplaces are all interesting. But in the grand scheme of things they feel like a side-show – tools to help customers amplify the value of Salesforce’s new operating model.
Because as I have previously argued, if agents are going to matter, they will not all sit inside the enterprise. They won’t stay bounded to a single system, a single organization, or even a single industry. They will roam freely and act on behalf of customers. They will reach into your infrastructure to facilitate partner actions.
Effectively they will integrate context across organizational boundaries in pursuit of their owners’ intent, rather than being constrained to the rigid workflows of a single enterprise.
Inzerillo was also explicit on this point:
I think one of the things about agents is they hate boundaries. Agents do not like to stay in the silo. They want to talk to each other. They want to talk to endpoints.
What means is that the challenge is no longer just to build agents that can navigate internal systems, but to make those systems accessible to agents that the enterprise does not necessarily control.
And Salesforce’s move is a concrete example of that idea in practice.
Because its platform is – in effect – also its business, changing how that platform is exposed is equivalent to changing how it operates. It is therefore a meaningful step in transforming its core operating model for an agentic era.
Headless 360 is, in that sense, Salesforce’s version of an operating model transformation – one in which it is refactoring itself so that its capabilities can be discovered and used as part of a broader ecosystem. It is signaling its belief that future upside will come from being an anchor within a wider ecosystem, rather than simply a provider of applications, positioning itself as a central hub through which agent-driven work can be assembled.
That has clear strategic logic.
Because if the interface layer fragments – as agents take on more of the work of navigating systems – then the value of applications declines, which represents a direct threat to an application-centric company. But it also shifts value toward companies that hold valuable context and action – but only if those capabilities can be accessed in the right form.
So Headless 360 looks to me like a fundamental move by Salesforce to make its capabilities legible – and in doing so cement its continued relevance in a world of agentic ecosystems.
For its customers, however, the company is also offering a valuable example of what it means to become an agentic enterprise – one which may be uncomfortable once they recognize it.
Because today most enterprises are not even thinking of re-configuring themselves in this way.
Their capabilities remain embedded in applications, distributed across teams, and often only partially defined. They rely on people to bridge the gaps – to interpret context, navigate systems and connect actions – and in that environment it might be easy to add agents as interface, but it is much harder to make agents part of the underlying infrastructure.
Which means the problem – once seen in this context – is less one of data and more one of operating debt. You can shuffle your data around as much as you want, but unless you make real capabilities – ones that make sense to agents – the foundation of your operating model, you will always struggle to adapt. Data quality has to be viewed through that lens, not in the abstract.
Which is why the most interesting thing about the first day of TDX for me was not the specific tools that were announced, but the practical demonstration of what Salesforce is doing to itself.
A shift which is showing, whether the audience fully recognizes it or not, what operating model transformation looks like during the move to an agentic era.
Not a shift toward more automation layered on top of existing systems, but a restructuring of those systems so that their underlying capabilities can be exposed, understood and invoked as part of a wider ecosystem.
My take
For other companies, the challenge is to work out what that means in their own context.
What are their discrete capabilities? Where do they sit? How are they defined? And how would they need to be exposed for an agent to use them?
Because if not all future agents will work for your enterprise, then your enterprise will need to be able to work with agents that are part of something larger than itself.
Which means the real heart of the transformation to the agentic enterprise isn’t, in fact, building agents.
It is becoming something they can actually use.