Interview with Notion CEO Ivan Zhao — custom Notion AI agents launching, agents build 50%+ of databases reports Notion launching custom AI agents that already build over half of Notion databases. This demonstrates agents moving from assistants to producers of production artifacts and forces outcome engineers to design for agent-led data creation, orchestration, and continuous validation (Principle 09, Principle 03).

How Tinfoil Proves Exactly What Model Is Running shows Modelwrap cryptographically binding published weights to a running server and using attestation to prove the exact model served. Outcome engineers gain a practical path to verifiable inference and supply-chain integrity—essential for audits, compliance, and trusting agent decisions (Principle 02, Principle 10).

NTransformer — Llama 3.1 70B on a single RTX 3090 via NVMe demonstrates streaming model layers via NVMe-to-GPU to run Llama 3.1 70B on a single RTX 3090. That lowers the infrastructure floor for running large agents locally, changing deployment trade-offs for latency, cost, and stateful agent workflows (Principle 07, Principle 12).

Anthropic data: software engineering accounts for ~50% of AI agent tool calls; remaining verticals are wide-open publishes usage data showing software engineering consumes about half of agent tool calls. Outcome engineers can use this map to prioritize product-market fit, intentionally expand into under-served verticals, and design generalizable agent interfaces (Principle 06, Principle 12).

Amazon: AI-assisted hacker breached 600 FortiGate firewalls in 5 weeks reports an AI-assisted attacker automating reconnaissance and credential extraction to compromise 600 FortiGate devices. This highlights that AI multiplies attack scale—outcome engineering teams must harden agent endpoints, manage secrets, and build an immune system for detection and recovery (Principle 14).