Link Dump #241

The cup of coffee or something to read? I'm glad this is the choice I do not have to make. 

1. Software Development
   1. Interrogatory LLM #PickOfTheWeek
      Are you spending hours trying to write the "perfect" context for your AI agent? Martin Fowler suggests a flip in the script: the Interrogatory LLM. By instructing the AI to interview you - one question at a time - you can extract the deep domain knowledge in your head without the pain of staring at a blank page. Learn how to turn your AI into an expert interviewer to build better feature specs and architectural reports.
   2. Goldratt, ToC, LLMs and the future of software
      Applying manufacturing principles to software development is nothing new, but LLMs are rewriting the rules. This deep dive into Eliyahu Goldratt's frameworks explores how agentic coding forces a shift in engineering focus.
   3. Slop Creep: The Great Enshittification of Software #PickOfTheWeek
      Bad code used to have a natural speed limit: it became too painful to work with, forcing teams to slow down and refactor. The author explains that AI agents have completely bypassed this roadblock, making it easier than ever to build massive piles of unvetted logic. Find out why the engineering role is shifting from typing lines of code to managing the "one-way doors" of data models and system boundaries.
   4. LLMs are functions, not brains
      Learn why the path to enterprise-grade AI infrastructure relies on stripping away the illusion of autonomy and treating LLMs as strongly-typed, stateless text-transformation functions.
   5. Migration Observability: Measure Meaning, Not Movement
      When we focus purely on data movement - like records copied or throughput - we miss "semantic drift," where the new system misinterprets the old data. Learn how to stop measuring progress and start measuring meaning before you pull the plug on your legacy infrastructure. 
   6. Production is the New Prototype: Notes from LangChain Interrupt 2026 #PickOfTheWeek
      If your team has been tinkering with localized AI demos for two years, you aren't building a strategy - you're just "agent washing." Discover the five architectural patterns separating enterprise successes from failures, and why real-world agent engineering requires putting observability and cost throttles ahead of your first deploy.
   7. CI/CD Was Built for Deterministic Software - Agents Just Broke the Model #PickOfTheWeek
      In the age of autonomous software, continuous delivery doesn't end when the artifact is healthy. The author explains that agents are systems capable of making independent choices post-deployment. If an agent triages an incident or opens a ticket incorrectly, restoring the previous commit won't fix the operational mess it left behind. Find out why the next generation of CI/CD must focus on behavioral guardrails and policy enforcement over basic code compilation.
      
2. Agile
   1. Story Splitting: How To Split User Stories So Teams Can Finish
      When a story is too large, the default engineering impulse is often to split it by technical layers - doing the database work now and the UI later. The author explains why this strategy destroys feedback loops. This deep dive breaks down how to use data variations, business rule relaxation, and path reduction to deliver fully functional, end-to-end features every single iteration.