If you run on Salesforce, you’ve probably already seen the highlights from World Tour. The keynote was slick, the demos were genuinely impressive, and there were AI agents booking and summarising and resolving things in a few clicks. By the time you walk out, you half believe you could go back to the office and have it all running by Friday.
A few of us were in the room in Melbourne last week, and we came away pretty excited about parts of it. We also came away with a clear sense of how much distance there still is between what was on stage and what most businesses can actually do with it today. So rather than give you another “everything announced at World Tour” rundown, we wanted to share the honest version: what we think is genuinely worth paying attention to, what we’d gently suggest you wait on, one message that wasn’t quite said out loud but clearly wanted everyone to hear.
The big idea: AI needs context, and Salesforce wants to own it
If there was a single word running through the whole event, it was context. The thinking is that AI is only ever as good as what it knows about your business, and the reason so many AI pilots quietly disappoint is that the model is basically guessing in the dark. Salesforce is positioning its own stack as the fix for that, with Data Cloud as the system of context, Customer 360 as the system of record, Tableau as the system of insight, Agentforce as the system of agency, and Slack as the system of engagement.
It’s a tidy story, and the bit underneath it is true. An AI agent that can’t see your customer data, your history, and your business rules is really just a chatbot with good manners. The agents that actually earn their keep are the ones grounded in real, governed data, so the core point holds up well. Whether getting there means you need every single layer of the Salesforce stack is a slightly different question, and one we’d encourage you to think through properly rather than just nod along to.
What’s genuinely worth paying attention to
The first thing is the shift from topics to sub-agents. Earlier versions of Agentforce had you building around “topics,” and the new model uses a top-level agent that hands work off to specialist sub-agents underneath it. That might sound like a small naming change, but it’s actually a much more sensible way to keep an agent precise and under control, because each sub-agent has one narrow job instead of a single agent trying to do everything and wandering off course. If you’ve been nervous about agents going off-script, this is probably the part of the redesign that should put you most at ease.
The testing side has come a long way too. Building agents is getting easier, but more importantly, testing them is getting more serious, with proper evaluation workflows, dashboards, and a library of ready-made agents to start from. One of our team who’s spent real hours in the platform put it well: the value isn’t really the natural-language demo, it’s being able to mix deterministic rules with natural-language behaviour. You get certainty where you need it and flexibility where you want it, and that combination is what makes an agent something you’d actually trust in front of a customer.
We were also glad to see governance pushed right to the front. Salesforce led with trust as its number one value, with security, permissions and governance built into the architecture rather than bolted on afterwards. For anyone in a regulated industry, that’s exactly the right emphasis. The harder questions tend to show up once agents start acting on your data on your behalf, because then you need to be able to prove what they did, why, and on whose authority, and that’s a governance challenge most businesses haven’t really sized up yet. The platform giving you the controls is the start of that answer, not the whole of it.
What we’d gently suggest you wait on
The headless 360 demo was probably the most jaw-dropping moment of the day, and it’s also the one we’d be most careful about. It showed a customer completing a purchase inside ChatGPT, with Salesforce quietly running everything behind the scenes. The idea, which they’re calling headless 360, is that the Salesforce interface becomes optional and the capabilities get distributed out to people, AI agents and other apps through APIs, all while keeping your governance rules and metadata intact. It’s a genuinely interesting direction to be heading in. It’s also a long way from something most customers can deploy today, so we’d treat it as a glimpse of where things are going rather than a feature you can switch on next quarter. Worth being excited about, but probably not worth building this year’s roadmap around just yet.
The message that wasn’t said out loud
This is the part that stuck with us the most. In one of the sessions, a retailer’s customer care team, not its developers, had built the solution being showcased on stage. The quiet implication, which came through in a few different forms across the day, was that you might not need a partner at all if your own people can learn the platform and build for the simpler use cases.
We’re not going to pretend that message wasn’t there, because it was, and honestly it’s partly true. For a small, contained use case, with the right person internally and a bit of time to learn, you can absolutely build it yourself. That’s not a new debate either. It’s the same build-versus-buy decision every business has always faced, and the right answer really does depend on your size, your industry, and how complex the thing actually is.
Where it stops being true is at scale. The moment a build starts touching integration, data quality, security and governance across a platform your whole business runs on, “we’ll just figure it out internally” has a habit of becoming the most expensive option of all, paid back in technical debt and rework a year or two down the track. So the honest framing isn’t “you always need a partner” or “you never do.” It’s worth knowing which kind of project you’ve actually got before you decide.
And honestly, if all a partner does is build the thing you could have built yourself, they’re not worth much. The value of a good partner these days isn’t the building. It’s stopping you from quietly piling up technical debt, helping you choose the use cases that will actually pay off, and making sure the money you put into Salesforce comes back to you. Think of it less as hiring a pair of hands and more as having someone in your corner whose job is to keep you out of the expensive mistakes and focused on the things that move the needle. That’s the relationship worth having, whether the build is big or small.
So what should you actually do with all this?
If you take one thing away from World Tour, we’d make it this. The technology is real and it’s improving quickly, but the distance between a keynote demo and a working agent inside your business is where most of the value gets won or lost. The businesses that get real value out of Agentforce won’t be the ones who move the fastest. They’ll be the ones who pick the right use case to start with, can show their data and governance are genuinely ready, take the time to scope the real investment to roll it out properly, and are honest with themselves about what they can build alone and what they can’t.
That’s the unglamorous bit nobody puts on stage, and it’s also the bit that decides whether any of this works for you.
If you’re weighing up Agentforce and want a straight read on whether your Salesforce is actually ready for it, that’s a conversation we’re always happy to have. No pitch, just an honest look at where you stand.
