Descriptions:
Recorded over the 2025 holiday season, this video from Nate B. Jones lays out a thesis about where real AI value will be generated in 2026: not from frontier model benchmarks or splashy demos, but from disciplined system design around smaller, faster, constrained models — what he calls “flash models.”
Jones argues the field is transitioning from LLMs as content generators to LLMs as software, and the teams that win will be those that invest in protocols and structured interfaces rather than clever prompting. He introduces an entropy framework for evaluating agentic system design: many 2025 pipelines accidentally increased entropy by stacking unconstrained steps and loops, while the highest-quality AI experiences — he names TLDraw and Figma’s Capsules feature alongside Claude Code and Codex — are ones where the LLM is tightly harnessed to produce coherent, lower-entropy outcomes for users. Getting that right requires constraints, structured outputs, reliable tool-calling, and defined handoff points between components.
He also highlights the talent profile he’s most optimistic about: individuals who can hold technical implementation details (verification loops, structured output schemas, tool permissions) and customer outcome thinking simultaneously — a skillset that bridges the gap between engineering and product. The video serves as a useful strategic orientation for teams planning 2026 AI system architecture and for anyone trying to understand why production agentic systems have underdelivered relative to demo-stage expectations.
📺 Source: AI News & Strategy Daily | Nate B Jones · Published December 24, 2025
🏷️ Format: Opinion Editorial
