Maribel Lopez reports live from AWS re:Invent 2025 in Las Vegas, unpacking why the AI experimentation phase is officially over. With statistics that say 95% of AI projects are failing and enterprise budgets tightening, 2026 demands production-quality AI—not more proof-of-concepts. This episode explores the critical shift from building agents to deploying them safely at scale.

Key Themes

The Reality Check (2025 Recap)

MIT study reveals 95% AI project failure rate
McKinsey and BCG document widespread implementation struggles
Board-level AI initiatives now demand real ROI, not just innovation theater
The POC gold rush is over—experimentation budgets are drying up

Agentic AI Grows Up The conversation has evolved from "can we build agents?" to "can we trust them in production?" Three critical roadblocks:

Security & Orchestration: How agents interact without creating vulnerabilities
Policy & Governance: Preventing rogue agents and establishing guardrails
Observability: Real-time monitoring to ensure agents perform as intended

AWS re:Invent 2025 Highlights

Agent Core Improvements

Enhanced policy frameworks defining agent boundaries and permissions
Human-in-the-loop controls for high-stakes decisions
Better cross-stack orchestration for multi-agent workflows

The Discoverability Problem

AWS Marketplace now features natural language search
Upload requirements documents instead of filling rigid forms
AI-suggested prompts help non-technical users navigate complex decisions
Smarter filtering for nuanced needs (performance vs. cost vs. compliance)

The Full-Stack Maturity

Recognition that AI "takes a village"—no single vendor owns the entire stack
Growing emphasis on open standards (A2A, MCP) for SaaS integration
Tools designed for all skill levels, not just data scientists

Key Takeaway

Enterprise AI in 2026 isn't about doing more—it's about doing it right. The winners will be organizations that prioritize governance, observability, and practical deployment over flashy demos.

Host: Maribel Lopez
Recorded: AWS re:Invent, Las Vegas, December 2025
Follow-up: Stay tuned for next week's deep-dive episode with demos and vendor interviews

From AI Chaos to Production: Why 2026 is the Year Enterprise AI Gets Real

The AI with Maribel Lopez (AI with ML)