Salesforce’s woes underline marketing’s agentic AI problems

Salesforce’s Agentforce Stumble Is a Mirror for Every Enterprise AI Ambition

Marc Benioff does not do quiet launches. When Salesforce unveiled Agentforce in late 2024, he declared the company “all in” and positioned autonomous AI agents as the next epoch of enterprise software — bigger than the cloud, bigger than mobile, possibly bigger than the internet itself. That is a very Benioff thing to say. The problem is that Wall Street, and more importantly customers, are now grading the exam, and the score is uncomfortable. Only 34% of Salesforce’s 150,000 customers have adopted Agentforce. The stock has shed more than 50% from its December 2024 peak, erasing north of $200 billion in market value. Two major analyst houses — KeyBanc and Bernstein — downgraded the stock on the same day, a rare convergence that signals something more than routine skepticism.

The instinctive read is that this is a Salesforce story. It is not. It is an enterprise AI story, and every C-suite executive currently signing off on an agentic AI roadmap should be reading it that way.

The Real Diagnosis: Data Debt Comes Before Agent Deployment

KeyBanc’s analysts, led by Jackson Ader, identified two failure modes in their downgrade report. The first is data readiness — AI agents require clean, structured, and connected data to make autonomous decisions, and most enterprises are still operating with fragmented CRM records, siloed systems, and inconsistent customer information accumulated over decades of organic and acquisitive growth. The second is product maturity — the majority of Agentforce deployments remain at the proof-of-concept stage rather than at enterprise-wide rollout scale. KeyBanc’s CIO survey found that more organizations expect to reduce Salesforce spending over the next twelve months than increase it. That is not a ringing endorsement of an “all in” strategic bet.

The blunt summary from KeyBanc deserves to be quoted directly, because it will resonate far beyond one vendor’s earnings cycle: “Customers’ data is not in order to do meaningful AI work,” and “Agentforce, as a product, just isn’t there.” Two separate problems. Both real. Neither easy to fix on a quarterly timeline.

Benioff pushed back hard, calling the KeyBanc report a “bad call” and citing internal metrics showing Agentforce as the fastest-growing product in company history. He is not entirely wrong to push back — Andreessen Horowitz data suggests that companies investing heavily in AI actually increased their median Salesforce spending by 3% over the prior quarter, and both Guggenheim and Monness, Crespi, Hardt upgraded the stock, seeing meaningful upside. The picture is genuinely mixed. But the bullish case and the bearish case are not actually in conflict: early AI adopters with clean data foundations are moving forward, while the broader enterprise population is stuck in data remediation purgatory. That bifurcation is the real story.

Why This Is Structurally Different From Prior Enterprise Software Cycles

Every major enterprise software platform has faced adoption friction. ERP rollouts in the 1990s were famously brutal. CRM adoption in the 2000s was chronically low until mobile made it unavoidable. Cloud migrations stretched across entire budget cycles. But agentic AI introduces a qualitatively different barrier that did not exist in those prior transitions.

Traditional enterprise software could be deployed on top of messy data because it was, at its core, a system of record and a workflow tool. A sales rep could manually correct a bad record. A finance team could reconcile a mis-categorized transaction. Human judgment patched the gaps. Agentic AI inverts that relationship entirely. The agent is the judgment layer. It does not patch gaps — it amplifies whatever signal it receives. Feed it fragmented, inconsistent, or stale customer data and it will act on that data autonomously, at scale, without a human in the loop to catch the error. The failure mode is not a bad report. It is a bad customer interaction delivered a thousand times before anyone notices the pattern.

This is why the data readiness problem is not a temporary implementation headache. It is a structural prerequisite. And it is why CIOs who rushed to procure agentic AI tooling in 2024 are now discovering that the procurement decision was the easy part. The hard part — data governance, integration architecture, master data management, identity resolution across systems — was already overdue and cannot be shortcut.

The CMO and CRO Implications Are Particularly Sharp

The marketing and revenue functions are exactly where the agentic AI promise is loudest and the data debt is deepest. Autonomous campaign execution, real-time lead qualification, personalized customer journeys, AI-driven customer service deflection — every one of these use cases depends on a unified, current, and accurate picture of the customer. Most marketing stacks are the opposite of that. They are a layered archaeology of point solutions, acquired platforms, agency-managed tools, and legacy integrations, each with its own data model and update cadence.

CMOs and CROs who are currently evaluating or deploying agentic AI tools need to ask a harder version of the question than most vendor conversations will prompt. It is not “can this agent automate our nurture sequences?” It is “what is the actual quality and completeness of the customer data this agent will act on, and what is the error rate we are willing to accept at autonomous scale?” If the honest answer is “we don’t know,” that is the answer that determines the deployment timeline — not the vendor’s product roadmap.

Salesforce’s own response to the adoption problem is instructive here. The company has been expanding its data ingestion capabilities, pulling customer data from external sources automatically, and accelerating data management ambitions through its Informatica acquisition. That is not a feature addition. It is an acknowledgment that the platform cannot deliver on its core promise without solving the data layer first. Every other agentic AI vendor is facing the same architectural reality, whether or not they are as publicly forthcoming about it.

The Strategic Reframe Every Executive Should Make Now

The Agentforce adoption rate is functioning as an inadvertent benchmark for enterprise AI readiness across the market. The companies moving through proof-of-concept into production are not, in the main, the ones who bought the most sophisticated AI tooling. They are the ones who spent the prior two to three years doing the unglamorous work of data unification, governance policy, and integration architecture. They had a foundation. They are now building on it.

The companies still stuck in proof-of-concept are not failing because the AI is bad or because their teams lack ambition. They are failing because the prerequisite investment was deferred and the bill is now due simultaneously with the AI adoption push. That is a resource allocation and sequencing problem, not a technology problem.

For CIOs and CDOs, the practical implication is to treat data readiness as the gating item on every agentic AI business case — not as a parallel workstream, but as the critical path. For CFOs approving AI budgets, the question worth asking is what percentage of the proposed spend is allocated to data infrastructure versus model and agent licensing, and whether that ratio reflects the actual constraint. For CEOs and boards being presented with agentic AI transformation narratives, the Salesforce moment is a useful stress test: if your own enterprise tried to deploy autonomous agents against your current CRM data today, would the outcome be transformation or embarrassment?

Benioff is almost certainly right that the long-run opportunity for agentic AI in the enterprise is enormous. He may also be right that KeyBanc’s downgrade is premature on a multi-year view. But the market is pricing the near-term reality, and the near-term reality is that most enterprises are not ready. The gap between agentic AI’s potential and enterprise AI readiness is not a vendor problem. It is an enterprise infrastructure problem that no amount of product marketing will resolve. The companies that close that gap first will capture the asymmetric returns. The rest will spend the next several years in an expensive, frustrating, and very public remediation cycle — with Salesforce’s stock chart as their cautionary wallpaper.

Based on reporting from Salesforce’s woes underline marketing’s agentic AI problems, originally published 2026-07-17 11:11:00.