Today at 2pm PT, @isaacbmiller1 and I will be walking through @DSPyOSS's Jev/System One implementation & the ReAnchor optimizer. Will also sharing some design patterns for for selective compaction, tool approvals, and subagent delegation using Jev.

As reasoning models consume more tokens and AI systems become more expensive to run, understanding what those tokens actually buy is becoming increasingly important. In this episode, @Stanford professor and Big Spin co-founder @ChrisGPotts joins us to discuss AI tokenomics and his research into “tokenflation”—the possibility that token usage is growing faster than the measurable value those tokens produce. We explore how to measure the return on AI spending, why benchmarks alone provide an incomplete picture of model progress, and what inference-time scaling means for the economics of increasingly capable models. Chris also explains why expert AI users tend to get better results by challenging and iterating with models, how AI fluency affects outcomes, and why more efficient architectures could change the underlying economics. We also discuss DSPy, interpretability, the limits of today’s transformer architectures, and where Chris sees opportunities for more fundamental innovation in AI. 🗒️ Full show notes: 📖 CHAPTERS =============================== 00:00 - Introduction 05:38 - Linguistics in the Age of Language Models 09:26 - Scale Limitations in NLP Research 12:54 - Challenging the Bitter Lesson Mindset 15:12 - Relationship Between Data, Mechanistic Interpretability, and Efficiency 17:03 - DSPy 21:32 - Prompt Optimization and Model Variability 24:35 - Tokenomics and the Rising Cost of AI 28:13 - Measuring Token Purchasing Power with a CPI 32:18 - Inference-Time Scaling 35:36 - Defining Value Across Different AI Tasks 38:33 - AI Value Creation 40:38 - Predicting AI Costs 42:25 - Tokenflation 46:22 - AI Fluency 50:12 - Key Lessons of AI Fluency Work 54:44 - Future Directions

.@hammer_mt starts a fresh chat for every task, even for the same project. He does this to avoid context rot, which happens when accumulated history makes AI responses worse. These are 7 tips for context engineering: 1. Start a new thread whenever the tone or type of task changes. 2. Keep memory turned off so irrelevant context can’t leak into the current conversation. 3. Only upload the documents the task needs. 4. Keep the preferences you always want—like spelling, tone, or length—in a saved document and paste them in at the top of every new chat. 5. Write the full brief before you start and put everything the model needs into the first message rather than drip-feeding it. 6. When the AI makes a mistake, go back and edit your previous message so it has no memory of doing anything wrong. 7. Before you leave a long thread, ask the model the question Mike has recommended when working on recurring tasks: What information would have been useful to know at the start of this task that would have eliminated several steps or mistakes? Get Mike’s new book, Context Engineering with DSPy: Read the full piece:

🧭 Reaching for LangGraph by default? This paper is a practitioner guide that separates where graph orchestration truly helps from where it just gets in the way. The authors argue LangGraph adds value not through model quality but through durable orchestration and governance. So it earns its place when a workflow must pause and resume, or when the next step depends on explicit state like risk, evidence quality, or retry count. Conversely, if a single prompt plus one tool call suffices, keep it simple with a plain SDK, they warn against overuse. The thesis is made concrete with three runnable recipes: SQL analytics that self-repairs from validation and execution errors; a fail-closed RAG that returns "insufficient" instead of fabricating when evidence is weak; and a HITL flow that interrupts only high-risk cases for human review. Each uses typed state, conditional edges, and checkpoints to lift repair and approval out of implicit prompts into testable, explicit steps (with a warning that removing the checkpointer breaks pause/resume, so high-risk cases never complete). Tool selection is crisp too: ReAct for simple cases, schema-first for structured extraction, DSPy for prompt optimization, LangGraph when branching and audit matter. By deliberately skipping benchmarks to discuss when engineering shape changes, Graph-Based Agentic AI with LangGraph is a solid map for design decisions. 🔗 #LangGraph# #AIAgents#