What Is Going On
Salesforce goes autonomous. Salesforce announced Agentforce with bold claims about autonomous customer service. The demo was impressive; the reality will be messier. Most enterprises can't even get their data clean enough for basic automation.
Microsoft focuses on reliability. Microsoft shipped a major Copilot update focused on 'grounding'—their term for reducing hallucinations in enterprise contexts. This is the real work: not making AI smarter, but making it reliable.
IBM finds its niche. IBM's watsonx governance tools are getting traction with regulated industries. Compliance is becoming a feature, not an afterthought.
AI Corner
Chain-of-thought doubts. New research on chain-of-thought prompting suggests structured reasoning helps less than we thought—models often 'fake' the reasoning to reach predetermined conclusions.
Open source fragments. The open-source model ecosystem is fragmenting. Llama derivatives now outnumber meaningful architectural innovations 10 to 1.
News You Can Use
- ●
Vertex AI upgrade. Google Cloud's Vertex AI added support for custom evaluation metrics. Essential for production deployments.
- ●
Notion opens up. Notion released an AI features API. Integration possibilities are interesting for workflow automation.
- ●
California vetoes safety bill. California's AI safety bill was vetoed. The regulatory landscape remains unclear.