Marc Benioff, who runs Salesforce, said in September 2025 that the company had cut roughly 4,000 support roles. He tied the cut directly to AI, saying he needed “less heads”. To leadership teams watching from outside, it looked like proof that AI was already reducing the need for support staff.
That is where the risk starts. A visible move by a large company can become a template for everyone else before anyone understands how it works in practice. Executives copy the technology, and they also copy the confidence behind the decision. When the timing is wrong, the costs add up fast.
AI can improve parts of a workflow without being ready to run all of it reliably. Customer support is a chain of conversations, context, policy decisions and exceptions. One weak link can wipe out the savings made by the stronger steps before it.
Salesforce’s own researchers put numbers on that weakness. Their CRMArena Pro study tested AI agents, software that carries out tasks by itself, across business scenarios. The best agents succeeded about 58% of the time on tasks settled in a single exchange. In longer conversations with several back-and-forth turns, success fell to roughly 35%. The agents also showed almost no built-in sense of which information was confidential and needed protecting.
Support work is full of follow-ups. An agent may draft a strong first reply and still fail when the customer changes the request, adds new information or asks for an exception. That work does not disappear. It returns as handoffs, retries and escalations, landing on teams that have already been cut.
Early gains are real. Routing requests, summarising and drafting replies can speed work up and reduce manual effort. Using those gains to justify permanent job cuts before the whole workflow performs reliably turns an efficiency project into a service risk.
The predictable sequel is quiet rehiring. The research firm Forrester expects about half of layoffs blamed on AI to be reversed, often by refilling the roles at lower pay. That blunts the savings and adds churn, while lost knowledge and damaged morale make service harder to stabilise. Forrester also found that 55% of the employers it surveyed regret layoffs made on the promise of AI.
