Everyday AI: Leveraging LLMs for Simple...
summary
Instacart demonstrates practical LLM automation: 95% auto-approve of identity requests using short prompts, Python Lambdas or a workflow builder (Tines + OpenAI) to make REST endpoints, a Slack-status LLM that launches in-production by lunch to detect approver availability, and Terraform-modifying automation that schedules quarterly runs to remove unused IAM permissions. Outputs are sanity-checked, PII is tokenized (Microsoft Presidio suggested), hallucination risk is acknowledged, and humans gate the final PR merge.
AI-written summary · about
chapters (16)
- 0:00Introduction and setup
- 1:03Speaker introductions
- 2:06Why use LLMs in security
- 3:50Getting started with LLMs
- 5:36New LLM capabilities
- 7:02AI expectations and limitations
- 8:10Identity security challenges
- 9:15Automating access approvals
- 10:40Slack status automation
- 11:07Cloud resource requests
- 13:15Policy as code changes
- 13:57Unused permission cleanup
- 15:01LLM context and PII
- 16:25Handling LLM hallucinations
- 17:06Human oversight in decisions
- 17:49Call to action
related talks
26mHow Bad Actors AbUse LLMs The Mechanisms That Enable Misuse and What We Can Do About
39mLeveraging AI and Emerging Technologies for Competitive Advantage
26mAI & SRE: Embracing the Ultimate Frontier of Automation
06mInterview of Marie Alice Blete
01h 39mMastering LLMs: From Prompt Engineering to Agentic AI
43mAI Development Lifecycle: Learnings of What Changed with LLMs
29mAutomating Privacy Reviews with LLMs: Challenges & Opportunities
29mGenerative-AI: Usecase-Specific Evaluation of LLM-powered Applications