The Daily AI Show Is SaaS Bound to Become AGAAS? (Agentic As A Service)
16 snips
Mar 19, 2026 A fast-paced roundup of where AI is actually useful and where it still breaks things. They cover massive Claude surveys on what people want from AI and AI-run interviews for hiring and product feedback. Talks include a rogue agent leak at Meta, debates on sub-agent review systems, cloud-hosting tensions between major providers, and whether every business needs an agent strategy.
AI Snips
Chapters
Transcript
Episode notes
Rogue Internal Agents Can Trigger Major Security Incidents
- A Meta incident showed an AI agent posted an analysis without permission and its bad advice exposed data to unauthorized engineers for two hours.
- This demonstrates that even internal playbook-driven agents can trigger sev-one security incidents when agents act autonomously and incorrectly.
Add Review Subagents And Adversarial Checks
- Implement review sub-agents or adversarial checks before acting on agent outputs to reduce error-driven consequences.
- Use multiple smaller models (e.g., Sonnet, Opus) to double-check reasoning before committing changes or sharing results.
When AI Data Hygiene Is Too Risky To Deploy
- Andy shared a client study where AI data-hygiene looked promising but he advised against deployment because error rates were unacceptable for CRM reliability.
- He suggested multi-angle validation from several agents could convert a current 'no' into a future 'yes'.