Memory Is Having a Moment. Tech & AI Weekly by Eli, August 14 2026
Two weeks of AI in one edition: context engineering from 35% to 98%, Meta's local 30B agent model, a shared plugin standard backed by five companies, and Anthropic's dream memory.
Two weeks of AI in one edition: context engineering from 35% to 98%, Meta's local 30B agent model, a shared plugin standard backed by five companies, and Anthropic's dream memory.
Stateless MCP lands, GPT-5.6 drops 80% in price, Claude Opus 5 ships, and 151 builders built 37 agent projects in one day.
Agent architecture over model choice: the context layer, converging agent platforms, Anthropic's $1.5B settlement, Kimi K3, and AI agent memory types.
Stripe benchmarked where agents fail, agents outspending humans, and the evidence-packet future of agent observability.
This week: a RAG pipeline that poisoned its own vector store, GitHub's AI agent tricked into leaking private repos, GPT-5.6, and what I'm digging into next: graph-based memory for agents.
Claude Fable 5 returns to Bedrock, Sonnet 5 lands on AWS, and AgentCore memory gets metadata filtering. This week in tech and AI.
This week: AgentJacking hijacks Claude Code, Cursor, and Codex through a fake Sentry bug report, new research on why models confuse trusted and untrusted text, Strands chaos testing, AWS Blocks, and Lambda MicroVMs.
This week: I published my book on AI agents that don't hallucinate or waste tokens, Strands cut agent costs in half, GLM-5.2 open weights, and the agent launches from AWS Summit New York.
Welcome to my very first newsletter!