1. Headline & intro
Salesforce is doing something many enterprise vendors talk about but rarely execute at scale: letting customers actively steer its AI roadmap, almost in real time. In an era where AI features can become obsolete in a quarter, this approach could be the difference between remaining a SaaS staple and becoming legacy software with a chatbot bolted on. In this piece, we’ll unpack what Salesforce is actually doing with its 18,000-strong customer feedback engine, why this matters in the AI platform wars, how it intersects with regulation and data risk, and what European organisations should take away from this experiment.
2. The news in brief
According to reporting by TechCrunch, Salesforce has built an unusually tight feedback loop between its AI teams and roughly 18,000 customers. Instead of traditional annual or quarterly advisory boards, Salesforce’s AI executives describe weekly or near‑weekly working sessions with some customers’ engineering and operations teams.
This collaboration helped drive the launch of Agentforce, Salesforce’s AI agent management platform introduced in late 2024, as well as a stream of new voice AI and Slack-related capabilities. Salesforce’s AI leadership says its roadmap is organised around themes such as context handling, observability and strong controls, rather than fixed multi‑year product plans.
Customer partners like travel platform Engine and credit union PenFed reportedly gain early access to experimental tools, while Salesforce observes how they are used, incorporates feedback and then rolls successful patterns back into the broader product portfolio.
3. Why this matters
For Salesforce, this is more than a customer‑friendly story; it is a survival strategy in a market where foundational AI models are rapidly commoditising. If everyone can plug into similar LLMs, differentiation shifts to three layers: data, workflow integration and trust. A live, high‑frequency feedback loop with thousands of enterprises gives Salesforce an edge on all three.
Who benefits?
- Salesforce narrows the gap between what it ships and what buyers will actually deploy at scale. It also deepens switching costs: if your workflows and even in‑house tools get productised into the core platform, you are far less likely to rip it out.
- Design‑partner customers get early access to AI capabilities plus the ability to shape them to real operational needs, not generic demos. For companies under intense competitive pressure, that time advantage matters.
Who loses?
- Smaller SaaS vendors that still work on 12–18 month roadmaps risk looking slow and tone‑deaf. If Salesforce can go from customer complaint to shipped feature in weeks, expectations for the rest of the industry rise.
- Some Salesforce customers may find themselves indirectly subsidising innovation for their competitors. When your bespoke workflow becomes a standard feature, your edge shrinks.
The deeper issue: Salesforce is betting that large customers actually know what they want from AI. Many don’t—yet. The risk is over‑indexing on short‑term requests from the loudest logos instead of building capabilities that help the broader market climb the AI learning curve.
4. The bigger picture
Salesforce’s move fits into a broader pattern in enterprise AI: the shift from one‑off features (“here’s a chatbot in your CRM”) to agentic platforms where AI systems act, not just answer. Microsoft is pushing the same logic with Copilot Studio and its orchestration tools; ServiceNow is doing it with Now Assist and workflow‑native agents; SAP is weaving Joule into its business suite.
What’s different here is the intensity and scale of co‑development. Design‑partner programmes are not new—cloud providers have long worked closely with a handful of lighthouse customers. What Salesforce is attempting looks closer to an ongoing, distributed R&D lab embedded inside 18,000 organisations.
Historically, big software vendors that stayed close to customers during platform shifts fared better. Microsoft’s early enterprise work around Office 365 and Azure created the foundation for Copilot. On the other hand, vendors that bet on static roadmaps during the mobile and cloud waves saw their relevance erode quickly.
AI agents amplify this effect. They are tightly coupled to business processes, data quality, security posture and regulation. You simply cannot design a credible agent framework in a vacuum. Salesforce’s weekly iterations with customers are, in that sense, a recognition that AI is no longer a feature layer—it’s operational infrastructure.
The message to the rest of the industry is clear: genAI is crossing the line from “innovation theatre” to day‑to‑day tooling, and the winners will look more like continuous co‑builders than distant software suppliers.
5. The European / regional angle
For European organisations, this approach is both appealing and complicated.
On the plus side, a vendor that is willing to sit down weekly with your engineers to tune AI agents to your real use cases is exactly what many EU enterprises have been asking for. Sectors like banking, insurance, public administration and manufacturing are heavily process‑driven; generic AI features don’t cut it. Co‑development promises better alignment with strict security requirements, on‑prem or EU‑only data residency, and detailed auditability.
But the regulatory context in Europe is far tougher. GDPR and the Digital Services Act already require strong controls around data use, logging and explainability. The EU AI Act—moving into enforcement over the next few years—will add risk‑based obligations for many AI systems used in credit scoring, HR, public services and beyond. If Salesforce is iterating weekly on agentic behaviours, European compliance teams will demand equally fast and transparent governance mechanisms: data‑processing agreements that cover experimental features, robust DPIAs for high‑risk use cases, and clear documentation of model behaviour.
European competitors like SAP, regional CRMs, and analytics players such as Celonis have an opportunity here. They understand EU regulatory culture deeply and can frame similar customer‑driven AI programmes with compliance‑first messaging. For mid‑size European SaaS companies, the lesson is not to copy Salesforce’s scale but to copy its cadence—frequent, structured feedback loops that translate directly into product decisions, with EU law embedded from day one.
6. Looking ahead
Over the next 12–24 months, expect Salesforce’s co‑development model to become more formalised and, ironically, more selective. You cannot meaningfully process weekly feedback from thousands of enterprises without hierarchy. We’ll likely see:
- Tiered customer councils for specific industries (financial services, public sector, manufacturing) that define agent patterns and guardrails.
- Standardised observability and control primitives—things like policy engines, audit trails and simulation sandboxes—that emerge directly from customer demands.
- Growing internal use of the same tools: Salesforce already says its own staff are heavy users, and that will increase as the company tries to find problems before customers do.
Risks remain. Customer fatigue is real; weekly calls that feel like unpaid consulting will not scale. There is also a strategic risk: if the AI infrastructure layer (models, vector databases, orchestration) standardises further, value will concentrate in a few horizontal platforms. Salesforce must prove that its agent framework is not just a nice UI over someone else’s stack but a defensible layer where customer‑specific learning accumulates.
For readers—especially those building or buying enterprise software—the key questions are: who inside your organisation is playing the role of "AI product manager" with your vendors, and how quickly can ideas from the frontline turn into shipped capabilities? Vendors that cannot answer that will slowly slide into irrelevance.
7. The bottom line
Salesforce’s AI strategy is a bet that tight customer co‑development will beat grand master plans in a market reshaped every few months. If it works, the company will turn its 18,000 participating customers into a distributed product organisation that competitors will struggle to match. If it fails, it will be because enterprises themselves are still unsure what they want from AI. Either way, the lesson is clear: in the age of AI agents, you can’t afford a passive relationship with your vendors. Are you in the room where the roadmap is being written—or on the mailing list that hears about it six months later?
