December 28, 2025
Bots benched, rules win
Salesforce pulls back from LLMs, pivots Agentforce to deterministic automation
Salesforce puts the bots on a leash; commenters cry 'obvious' and ask about 4,000 jobs
TLDR: Salesforce is dialing back chatty AI and pushing rule-based automation after reliability flops, cutting ~4,000 support roles. Comments swing between “we told you so,” PR-speak jokes, and job-loss worries—showing big companies want guaranteed actions, not easily distracted bots.
Salesforce is putting the chatty AI on a leash, pivoting Agentforce from big “guessing” models (LLMs—think very smart autocomplete) to predictable, rule-based automation. Why? In the real world, bots skipped steps after more than eight instructions and drifted off-task. One customer, Vivint, even watched post‑chat surveys go missing until deterministic triggers guaranteed they were sent. Execs now preach data-first guardrails over vibes, and the company quietly shrank support roles from ~9,000 to ~5,000.
The comments are pure popcorn. chrisjj snarks that “unreliable” got rebranded as “probabalistic”, while PartiallyTyped rolls eyes: it’s just the same “obvious” checklist folks have repeated for years. david_shi jokes the PR line — “Trusted. Reliable. Secure.” — sounds copy‑pasted by an LLM. Meanwhile, wood_spirit asks the blunt question: what about those 4,000 jobs? One spicy take was [flagged], but the vibe is split: some cheer a grown‑up move (‘please do the thing’ every time), others call it face‑saving theater and meme it as Ruleforce. Whatever you call it, the community agrees on one thing: mission‑critical work needs guarantees, not a chatbot that gets distracted by random questions.
Key Points
- •Salesforce is shifting Agentforce from heavy reliance on LLMs to deterministic automation due to reliability issues in production.
- •Executives reported models skipped steps when given more than eight instructions, undermining policy-bound workflows.
- •Vivint’s use case showed satisfaction surveys were inconsistently sent; deterministic triggers fixed the issue.
- •Salesforce reportedly reduced support roles from ~9,000 to ~5,000 after deploying AI agents, focusing on predictable outcomes and guardrails.
- •Leadership emphasizes data-first foundations and governed systems, with guidance on process design, instruction ceilings, guardrails, drift instrumentation, deterministic actions, and human-in-the-loop.