Salesforce has spent much of its existence convincing its customer base that its products work better together. Start with CRM, add Slack, MuleSoft, Tableau, Data 360, and now Agentforce, and you end up with a platform that covers a huge surface area of the business technology footprint. That joined-up experience has always been one of Salesforce’s biggest strengths, but it also makes the platform incredibly “sticky”, if you will.
The wider tech market, however, seems to be moving in another direction. More businesses want more choice over where their data sits, which AI models they use, and which tools make up their stack. Concerns around vendor lock-in are also rising, with recent Open Source Initiative research suggesting it is becoming a bigger reason companies explore open-source software.
Salesforce hasn’t looked past this, continually supporting external data platforms, different AI models, third-party tools, and generally finding more open ways of connecting systems across this year. But it does feel like there’s an interesting contradiction – Salesforce may be becoming more open while also becoming more central to how everything really fits together. So has Salesforce really reduced the risk of lock-in, or is that dependency simply changing shape as we speak?
The Modern Salesforce Stack Can’t Just Be Salesforce
The idea of being “locked into Salesforce” can make it sound as though customers are simply buying more and more Salesforce products until there’s nothing left. But the reality of the modern Salesforce stack is much messier than that.
A company may use Salesforce as its CRM, but Snowflake or Databricks for data, Power BI for analytics, a third-party DevOps platform, and any number of specialist tools around it. AI naturally broadens this too, with businesses having so many options or models and tools – OpenAI, Anthropic, Google, and others, which means they don’t need to rely on one provider.
Insights from the 2026 SF Ben Architect Survey show that Salesforce Architects spend an average of 45.1% of their time working with technologies outside Salesforce products. As we recently explored on SF Ben, the role itself is becoming less Salesforce-specific as architects are expected to understand larger tech stacks and systems beyond the platform.
This can also be seen in other individual areas of the ecosystem. Our look at Salesforce DevOps earlier this year found that third-party vendors and open-source CI pipelines remain widely used by developers and architects, while adoption of Salesforce’s own DevOps Center was much lower.
Now none of this looks like an ecosystem where customers are forced to use Salesforce for everything, but that doesn’t necessarily settle the lock-in question. If anything, it makes me want to query this – does Salesforce actually need to own every part of the stack to make itself difficult to replace?
Tim Combridge, Technical Content Writer at SF Ben, believes the traditional Salesforce lock-in argument is becoming weaker: “Does Sales Cloud, Service Cloud, etc. lock you in? I’d say it’s not as tight as it was a few years ago,” he explained. “Salesforce handles all the infrastructure that people just don’t think about, and offers a good application layer solution on top. But that’s easier to replace these days.”
Individual Salesforce applications may be becoming easier to replace or work around, but that doesn’t necessarily mean Salesforce itself is becoming less strategically important.
Salesforce Is Opening Up While Moving Further Into the Middle
This concept may sound a little confusing, so stick with me. Salesforce seems well aware that customers don’t really want to hand one company every part of their tech stack. In fact, some of its biggest announcements this year have been about making the platform easier to use alongside other technology.
Headless 360 is the clearest example of this. Announced at TrailblazerDX in April, the idea is that Salesforce data, workflows, and other capabilities can be accessed through APIs and newer tools like MCP, rather than requiring someone to actually sit inside the traditional Salesforce interface. As we covered at the time, Salesforce’s pitch was that developers and AI agents should be able to build on the platform “any way you want”.
There are also similar signs elsewhere, with Data 360’s Zero Copy capabilities allowing businesses to work with data stored in platforms such as Snowflake and Databricks without first moving all of it into Salesforce. Alongside that, Salesforce is becoming very comfortable with outside AI providers. As we’ve covered, Claude can now operate inside Slack, while Agentforce itself supports models beyond Salesforce’s own technology. And that’s to say nothing of their pre-Dreamforce announcement of an even closer partnership with Anthropic, now dubbed “Claudeforce”.
On the face of it, this might not sound “locked down”, but that is where things get interesting. Salesforce may not need to own the database, the AI model, or even the screen employees are working from. It becomes the layer connecting all of them.
We explored a version of this recently when Salesforce suggested users may eventually not even need to log into Salesforce at all, with Slackbot able to access CRM, Data 360, Tableau, and Agentforce through a new connection. That sounds more open, but it also potentially makes Salesforce much harder to remove.
So the reality is that by making itself easier to connect to everything else, it is putting itself right in the middle of everything instead. Quite the paradox, indeed.
Salesforce CTA Joshua Freeman sees some level of dependency as unavoidable when companies choose the platforms that will underpin their AI architecture.
“Any platform you choose to build your agents on, or to serve as your data orchestration layer, comes with some inherent dependency,” Josh told SF Ben.
However, he doesn’t think that automatically means Salesforce needs customers to buy its entire stack. In fact, he believes one of Salesforce’s biggest opportunities could be allowing Data 360 to succeed more independently.
How AI Changes What Salesforce Lock-In Actually Means
Vendor lock-in is hardly a new phenomenon for Salesforce customers. Years of customizations, integrations, trained employees, and business processes can make moving away from the platform a huge undertaking for everyone involved. AI doesn’t make that disappear by any stretch, but it could change where the dependency sits.
The AI market is moving incredibly quickly. Claude, ChatGPT, and Gemini are competing fiercely for enterprise customers, and there is no guarantee that today’s leading model will still be the obvious choice in a few years time. I explored recently that many Salesforce customers are effectively choosing between several paths when it comes to AI – invest heavily in Agentforce, buy a specialist AI product, or build more of their own stack around models such as Claude and OpenAI.
Salesforce Architects have told us they’re certainly not sticking to just one option, with our latest survey highlighting that ChatGPT, Claude, Gemini and Agentforce are all being widely used, with overall AI adoption among architects reaching 96.7%.
If the models themselves become relatively easy to swap, however, the harder thing to replace may be everything around them. And for Josh, that is where the lock-in conversation has to move.
“Wherever your business logic lives,” he said when asked where the real lock-in sits in an AI-driven stack. “AI has made moving data between systems much easier, but moving business logic and workflows is still a major undertaking.”
This also isn’t a problem that only Salesforce is having. Joshua points out that AI vendors themselves are also building broader platforms around their models, partly because owning more of the surrounding infrastructure makes customers less likely to leave.
“The model providers know this, which is why they’re building their own tooling and platforms, like OpenAI Frontier, to become stickier,” Josh explained.
For many Salesforce customers, the platform already knows who their customers are, what employees can access, which workflows should run, and where information needs to go next. If AI continues to rely on that context to actually do useful work, Salesforce could become more important even when the intelligence itself comes from somebody else.
That could also help explain why Agentforce has become so crucial to Salesforce’s overall strategy. Tim wonders whether Agentforce is partly Salesforce’s answer to the risk that increasingly capable external AI platforms could weaken its position.
He said: “Salesforce knows for a fact that AI is going to play a BIG role in the future, and so they’re trying to balance adding value with ensuring it keeps their core platform customers.”
Tim also noted the fact that Agentforce is fundamentally tied to the Salesforce ecosystem, whereas businesses using Salesforce, Microsoft Dynamics, or other platforms can potentially build around models and tools from companies like Anthropic and OpenAI – which is a hugely important difference.
“Where they’ll win is through Agentforce, if they get it right,” Tim added.
There is at least some concern about this already. 27.5% of respondents identified vendor lock-in as a concern when thinking about their Salesforce data foundations for AI, per our Architect Survey.
So perhaps AI isn’t removing the old lock-in question. It is just giving us a new version of it: if you can change the AI model but not easily change the system supplying everything that model needs, how much freedom do you really have?
Final Thoughts
It would be easy to look at all of this and come away thinking Salesforce is trying to have it both ways. In some respects, it probably is. Customers want more freedom, but they also want fewer headaches, and those two things do not always sit comfortably together.
Salesforce’s response has been to make itself easier to plug into, rather than insisting everything lives inside its own walls. That feels like a sensible adjustment to where enterprise technology is heading.
The real test will be whether customers continue to feel they have genuine choice, or whether Salesforce simply becomes so useful in the middle that leaving it becomes harder than ever.
