Salesforce announced general availability this month of ChatGPT integration for Agentforce Commerce, letting customers transact through OpenAI's platform using Salesforce's commerce agent infrastructure underneath — a notable move for a company that has otherwise positioned Agentforce as a walled ecosystem built around its own Atlas Reasoning Engine. It's the latest move in what's become an increasingly direct fight between Salesforce and ServiceNow over a genuinely new question: which platform becomes the default operating system that enterprise AI agents run on, the way Windows or a cloud hyperscaler became the default operating layer for the applications that came before them. For IT leaders currently evaluating either platform, or already committed to one, this is worth understanding as a platform-level bet, not a feature comparison.
What each company is actually building toward
Salesforce's approach extends outward from its CRM and customer experience roots. Agentforce's core pitch is autonomous agents handling customer-facing work — sales follow-up, service resolution, and now, with the ChatGPT integration, commerce transactions that can be initiated from outside Salesforce's own interface entirely. The company has been aggressively adding capability at a rapid pace: multi-agent orchestration letting agents collaborate on end-to-end workflows, support for Google's Gemini as an additional reasoning model alongside OpenAI and Anthropic on Amazon Bedrock, Agentforce Voice for natural conversational interfaces, and more than 50 specialized agents deployable out of the box across Slack, Teams, and IT service desk contexts.
ServiceNow is building from the opposite direction — its roots are in IT service management and back-office workflow automation, and its AI agent push leans into that strength: deep, complex workflow automation for internal operations rather than customer-facing interactions. Where Salesforce leads in customer-facing agent capability, ServiceNow's advantage is workflow depth for the kind of multi-step, cross-department internal processes — IT ticket routing, HR case management, procurement approvals — that most organizations run on ServiceNow's platform already.
Why "who owns the agent OS" is the right way to frame this, not "which agent is better"
The individual agent-versus-agent comparisons that dominate most vendor content miss the more consequential structural question. Both companies are explicitly positioning themselves as the platform other AI agents plug into, not just a vendor of agents themselves. Salesforce integrating ChatGPT into Agentforce Commerce, and supporting Gemini alongside OpenAI and Anthropic models in its Atlas Reasoning Engine, signals a deliberate multi-model strategy: Salesforce doesn't need to win the underlying model war, because it's betting its value sits in the orchestration and workflow layer above whichever model happens to be running underneath at any given moment. ServiceNow is making a structurally similar bet from its own workflow-depth position. Whoever wins this framing — becoming the layer enterprises build their AI agent strategy around, rather than one of several point-solution vendors — captures outsized value regardless of which underlying model providers ultimately dominate.
The real strategic risk for IT buyers: platform lock-in at a new layer
Enterprise software has been through this pattern before — CRM, then cloud infrastructure, then workflow automation each went through a consolidation phase where a small number of platforms became the default layer everything else plugged into, and switching away from the winner became increasingly costly over time as more of an organization's processes became entangled with that platform's specific data model and integrations. An enterprise AI agent operating system is a plausible next iteration of that same consolidation dynamic, and picking a side early — as many organizations are effectively doing right now simply by extending their existing Salesforce or ServiceNow relationship into agent territory — carries the same lock-in risk that CRM and cloud platform decisions carried a decade ago, just with considerably less historical precedent to draw on for how costly switching will eventually be.
What this means in practice for organizations running both platforms
A meaningful number of large enterprises run both Salesforce and ServiceNow today, for genuinely different purposes — Salesforce for customer-facing CRM and sales, ServiceNow for internal IT and operations workflows. As both platforms push AI agents deeper into their respective domains, the practical question shifts from "which platform do we choose" to "how do we prevent two separate agent ecosystems from creating duplicated logic, conflicting data ownership, and confusing handoffs when a customer-facing agent needs information that lives in an internal ServiceNow workflow, or vice versa." That integration question is arguably a bigger near-term operational risk for most IT organizations than the longer-term "which platform wins" strategic question, and it deserves its own architecture review independent of vendor marketing from either company.
Why Agentforce Commerce's ChatGPT integration is the specific detail that matters most
It's worth dwelling on why the ChatGPT integration into Agentforce Commerce is a more significant signal than it might first appear. Salesforce could have built Agentforce as a closed system where its own reasoning engine is the only path to transacting through its commerce agents — that would be the more conventional platform-lock-in play, maximizing switching costs by keeping customers dependent on Salesforce's own AI stack end to end. Instead, the company explicitly opened a path for commerce transactions to be initiated from a competitor's consumer-facing AI product. That's a bet that owning the underlying commerce infrastructure and workflow logic matters more than owning the conversational interface layer, and it's a meaningfully different strategic posture than, for instance, Apple's historically tighter approach to its own ecosystem integrations. Whether this openness is a genuine long-term philosophy or an early-stage move Salesforce reverses once it has enough market share to enforce more lock-in is exactly the kind of question worth watching over the next several quarters, since the answer materially affects how much platform risk customers are actually taking on today.
What to actually do about this as an IT or platform strategy leader
-
Map which of your business processes are already touched by AI agents on either platform, and where the handoff points between them sit. If Salesforce agents and ServiceNow agents both touch the same customer or employee record at different points in a process, that seam is where confusion, duplicated work, or data inconsistency will surface first.
-
Push both vendors explicitly on interoperability, not just capability, in your next renewal or expansion conversation. As both platforms build out agent ecosystems, their willingness to interoperate cleanly with each other — versus quietly making cross-platform integration harder to encourage consolidation onto one — is a legitimate, specific question to ask account teams directly.
-
Avoid over-committing organizational process design to either platform's specific agent framework this year. Given how early and fast-moving this space is, building critical business logic in a way that's tightly coupled to one vendor's current agent architecture risks a costly rebuild if either platform's approach shifts significantly, which is a real possibility this early in a genuinely new product category.
-
Track the multi-model support question closely for whichever platform you use. Salesforce's willingness to support ChatGPT, Gemini, and Anthropic models simultaneously is a meaningfully different bet than a single-model-locked platform, and it directly affects your own future flexibility if the underlying model landscape keeps shifting as quickly as it has this year.
-
Treat this as a multi-year strategic decision, not a feature bake-off. The vendor evaluation criteria that matter for "which chatbot is smarter" don't capture the actual stakes here — the more important questions are about data architecture, workflow ownership, and platform lock-in risk five years out, not which agent demo looked more impressive this quarter.
A practical test for evaluating either vendor's roadmap claims
Given how quickly both companies are shipping new agent capability announcements, IT buyers need a simple filter for separating durable platform investment from feature-of-the-month marketing. A reasonable test: ask whether a given announcement changes the underlying data model, integration architecture, or orchestration layer, versus simply adding another named agent to an existing framework. Multi-model support and the ChatGPT integration both pass that test — they change what the platform can structurally do. A new named vertical agent for a specific industry use case, by contrast, is usually a much shallower announcement dressed up with the same "AI agent" branding. Learning to tell the two apart quickly will save your evaluation team a meaningful amount of time sorting genuine platform signal from routine product marketing over the next several product release cycles from either vendor.
The bigger picture
This platform war is unfolding against a backdrop of intense multi-agent orchestration investment across the entire enterprise software industry, and Salesforce and ServiceNow are far from the only players — Microsoft's Copilot ecosystem and a wave of standalone agent-orchestration startups are pursuing similar platform ambitions from different starting points. But the Salesforce-ServiceNow rivalry specifically matters because both companies already sit deeply embedded in most large enterprises' existing operational data and workflows, which gives either one a genuine structural advantage in becoming the agent layer, rather than needing to win a green-field adoption decision the way a newer entrant would. Whichever way this consolidates, the organizations that come out ahead will be the ones that made deliberate, informed platform bets now — not the ones that let vendor account teams make the decision for them by default.