STUPID-2026-0062

Salesforce Agentforce hit a 77% B2B failure rate — and Salesforce admitted it was 'more confident than we should have been'

3.3low
June 1, 2026Verified

  1. Instruction given

    Deploy Agentforce AI agents to handle enterprise support and service workflows.

  2. Expected behavior

    Reliably automate complex enterprise workflows at scale.

  3. Actual behavior

    An Oliv.ai 2025 enterprise survey put the B2B Agentforce failure rate at 77%, driven mostly by data quality and Data Cloud lock-in; only 31% of deployments survived past six months, and 40% of agentic AI projects were canceled or paused by February 2026. Salesforce, which had replaced 4,000+ support roles with agents, later conceded it was 'more confident about AI than we should have been.'

  4. Damage

    Widespread enterprise failure to deploy reliably: most Agentforce deployments failed or were abandoned within months, after headcount had already been cut on the assumption they would work.

Salesforce's Agentforce became a case study in the gap between AI-agent marketing and enterprise reality. An Oliv.ai 2025 survey found a 77% B2B failure rate, driven mostly by data quality and Data Cloud lock-in — deploying agents on top of systems that worked 'well enough' for humans but couldn't survive being reasoned over at scale. Only 31% of deployments survived past six months, and by February 2026, 40% of agentic AI projects had been canceled or paused. The stakes were real: in early 2025 Salesforce's CEO announced agents had replaced the work of over 4,000 support and service employees — and by mid-2026 executives publicly conceded, 'We were more confident about AI than we should have been. The technology is powerful, but deploying it reliably in complex enterprise workflows is significantly harder than we anticipated.' It is the enterprise-scale version of confidence miscalibration — not one agent's bad action, but an organization betting on reliability that wasn't there.

Classification

Agent
Salesforce Agentforce
Failure mode
Other
Root cause
Confidence Miscalibration
Domain
Backend
Source
Benchmark

Related incidents

3.3
Cyera study: 344 verified enterprise agent-damage cases, 188 with no attacker involved
3.3
In independent testing, Devin completed just 3 of 20 real-world tasks (15%)
2.7
Anthropic found Claude Opus 4 would blackmail testers in up to 96% of simulated shutdown scenarios
2.2
The runaway-cost pattern, quantified: agentic coding tools burn 10-100x more tokens and can rival developer pay
Discuss on Hacker News

Get told when an agent breaks something

We document AI agent failures daily, severity-scored against a published scale. When one lands at 7.0 or above — deleted data, leaked secrets, broken production — you get an email with the source. When nothing does, you get nothing.

This database is callable over MCP — query it from inside your agent.