10.01.26 By Scott Effler
We went into Dreamforce this year with a running list of questions. Now we’re back. Some of those questions got real answers. A couple didn’t – and honestly, the ones still open are more interesting than the ones that got resolved. Here’s the scorecard.
The answer seems to be faster than the skeptics expect, slower than the keynotes imply – which is about where we landed.
Salesforce’s product teams are clearly doubling down here: inventing and shipping skills, and building UXs designed from the start to live inside other interfaces rather than assuming everyone logs directly into Salesforce. As customers’ own platform investments mature, those tool sets will increasingly plug into AIforce to surface Salesforce data and actions elsewhere.
Worth keeping in perspective: AIforce, the rebranding of Headless, is still a relatively new technology. It was announced at TDX earlier this year, which means it’s been in the world for less than twelve months. What we’re seeing now is early adopters starting to figure out how the pieces fit together. As more of those pieces land, we expect more companies to invest and start meeting their users where they already are – but “early” is the honest word for where this stands today.
The answer here isn’t a brand-new use case – it’s Salesforce leaning on what it has done best for two decades.
Salesforce’s own rollout playbook, laid out at Dreamforce, is deliberately unglamorous: pick a well-defined use case, deploy it fast, then customize and extend it, then observe and optimize once it’s live – not build something bespoke from scratch.
The skills and the MCP fabric that are being released first, aren’t exotic. They’re standard, common sales and service tasks: account management, opportunity management, case management, and giving reps and service teams a clearer, faster read on the customers and clients they’re already responsible for. This is “grandfather cloud” territory Salesforce has owned since before anyone was talking about agents – it is now happening from applications across the enterprise not just from Salesforce.
The published results back this up. On the service side, companies are reporting automatic case triage and routing to the right human agent, first-contact resolution for roughly 37% of inquiries with no person involved, and knowledge-management agents keeping answers consistent across channels. On the sales side, the headless capability lets account and opportunity intelligence surface directly inside tools reps already use (Slack, for instance), instead of requiring them to open Salesforce first to get the same read on a customer.
None of it requires customers to rethink their process – it’s the same job, done with less manual digging, using infrastructure and data Salesforce already owns. That’s exactly why it’s low-hanging fruit: fast to deploy, low complexity for the customer, and built on decades of Salesforce’s own account, opportunity, and case data rather than something new. If you’re looking for where agent adoption is real today vs. theoretical, this is it.
We’re still not sure – and we think that’s the honest answer.
The technology itself is legitimately impressive. Being able to reach Salesforce data from a different interface, surface it inside another AI platform, and gain insight without living inside Salesforce all day is a real capability, not vaporware.
But technology adoption was never really the hard part of enterprise software. Humans are creatures of habit, and asking someone who’s spent years perfecting a process, or who simply isn’t comfortable with AI yet, to change how they work is a genuinely hard ask, no matter how good the tool is.
There’s a silver lining hiding in that friction: processes people already find arduous or actively dislike doing are exactly the ones where new tooling has the best shot at driving real adoption. Nobody defends a process they hate. We’re also watching a generational split emerge – people who grew up using tech from the time they were toddlers are picking this up faster than people who didn’t, which isn’t surprising but worth tracking.
So, here’s what we still can’t tell: how much of the adoption curve we’re seeing is about the technology being ready, versus users digging in and staying inside the Salesforce UI they’ve always known, technology be damned. That’s going to take more time and more data than one Dreamforce cycle to sort out.
We made a point of hearing this directly from customers rather than just watching the keynote stage, and the honest picture is more mixed than the highlight reel suggests. The excitement is real – people are genuinely energized by what agents can do once they’re live. But that enthusiasm sits right next to real hesitation, and the two were present in almost every conversation we had.
Technical debt is the thread that ran through nearly all of it. Customers aren’t short on desire for the new capabilities; they’re boxed in by what they’ve already built – years of customizations, integrations, and processes that assume Salesforce works the way it’s always worked. Layering an agent on top of that isn’t a checkbox, it’s an unwind-and-rebuild problem for a lot of teams, and that reality doesn’t show up in a keynote demo.
The numbers back up what we heard in person: roughly 20% of agents that go into pilot across the industry never make it to production. The ones that do tend to share the same pattern – someone scoped the right first use case and defined KPIs before flipping anything on, rather than standing up an agent in a few minutes and hoping for the best. We heard both versions of this story, good and bad, from customers and from Salesforce, often within the same conversation.
So does the on-the-ground sentiment match the keynote narrative? Partly. The excitement is genuine, but it comes with a caveat the stage doesn’t spend much time on: the technical debt has to get cleaned up before agents can be trusted to run against it. That’s the honest balance – real momentum, paired with real work sitting underneath it.
This ties directly back to the technology-versus-human question, and it’s where we got the most useful reframe of the week.
There’s been talk for a long time about “removing the swivel chair”, the idea of eliminating the back-and-forth between systems. The realization we walked away with is that Salesforce was never really the only chair. The actual advantage is being present inside whatever application someone already lives in for their work. Some people are going to live inside Claude Code, some inside Salesforce itself, some inside their inbox or Teams. Meeting people where they are isn’t a slogan; it’s the actual mechanism that makes this useful.
AIforce is doing the work of surfacing information inside whichever application someone is using, for part of or all of their day. Marc Bennioff noted in the keynote that the treasure trove within Salesforce lies in the data that can be unlocked using AIforce within other platforms. As all these capabilities GA, we’re watching for how the technology will scale across processes and organizations, which is likely a story we’ll be able to tell more fully by next year’s Dreamforce.
Of the five questions we walked in with, a few got concrete answers: where the low-hanging fruit is, roughly how far along HXL really is, and the swivel-chair reframe on “work from wherever you are.” Two are still genuinely open: whether this is fundamentally a technology problem or a human one, and how the customer-adoption curve settles as more agents move from pilot to production.
If anything, the open questions matter more than the answered ones, they’re the ones that will determine whether this platform shift lands the way the keynotes suggest, or if it will take longer than anyone on stage wants to admit. We’ll be watching both, and if you’re working through similar questions on headless timelines, agent adoption, cleaning up tech debt, or scoping your first agent the right way, we’d welcome the chance to compare notes.
