Writing / 35 dispatches

Notes from the work,
not the sidelines.

Daily insights on rolling out AI at large firms.

  1. 01 Build the Data Platform Once A field guide to building a governed customer data platform from collection through activation, with an architecture ready for AI agents.
  2. 02 Agent Skills: packaging what your org knows An Apache licensed spec for portable agent capabilities. The enterprise angle is that know-how stops being locked to one vendor's agent.
  3. 03 Open weights walk in the front door of GitHub Copilot Kimi K2.7 Code is the first open weight model in Copilot's picker. The story is distribution, not benchmarks.
  4. 04 Kimi K3 changes your leverage, not your stack The first open 3T-class model is here. Almost no enterprise will ever host it, and that is not where the value is.
  5. 05 MiniMax M3: judge it by the 24 hour run MiniMax's new flagship pairs a 1M token context with the metric that matters for agentic delivery: how long it stays coherent without a human.
  6. 06 Agents Need Eyes: The Verification Layer Is Arriving Chrome ships DevTools as an MCP server for coding agents. It is one of the clearest signals that verification, not code generation, is where agentic delivery gets decided.
  7. 07 Astryx: Meta Makes the Design System the Agent Contract Meta's new open source design system calls itself agent ready. That phrase is the strategy: components as the guardrails agents build UI through.
  8. 08 HubSpot AEO: An Incumbent Prices AI Search Visibility at $50 HubSpot now sells answer engine optimization as a standalone product. When an incumbent productizes a category at $50 a month, the category stops being exotic.
  9. 09 TrueDoc: Document Fraud Detection for the Generative Era AI made convincing fake paystubs free to produce. TrueDoc is the countermeasure: ensemble detection for forged and AI-generated documents at intake.
  10. 10 Bringing a Regional Insurer Into the AI Era: Start With the Gateway, Not the Chatbot A hypothetical first 90 days at a regional insurance carrier. The right first move is a model gateway and shadow AI measurement, not a customer-facing chatbot.
  11. 11 OpenAI Shipped a Codex Plugin for Claude Code. Read the Signal. The official codex-plugin-cc repo lets Claude Code users run Codex reviews and delegate tasks. The interop matters more than the feature.
  12. 12 Graphify Turns Your Repo Into a Knowledge Graph Agents Can Query An open source skill that maps code, docs, and media into a queryable graph for coding agents. Context retrieval is the real constraint on agent quality.
  13. 13 TestMu's Browser Cloud Sells Real Browsers to AI Agents The company formerly known as LambdaTest now offers managed Chrome sessions as agent infrastructure. Verification capacity is becoming a line item.
  14. 14 AI for customers: ship the boring capability first The first layer of enterprise AI is customer-facing product. Most companies get it backwards by shipping a chatbot before removing a single point of friction.
  15. 15 AI for employees: shadow AI is a demand signal, not a crime wave The second layer of enterprise AI is workforce enablement. The companies that win it treat unsanctioned AI use as product feedback for the internal platform.
  16. 16 Articos and the synthetic user research bet AI user research in about 30 minutes with synthetic interviews and no recruitment. Fast and cheap, if you understand exactly what it cannot tell you.
  17. 17 A joke repo just made the best point about token economics Caveman, a Claude Code skill that cuts about 65 percent of tokens by talking like a caveman, went viral. The punchline is a real budget line.
  18. 18 AI as the engineering factory: copilots are stage one, not the strategy The third layer of enterprise AI redesigns how software gets built. A maturity model for going from individual assistants to an agentic delivery system.
  19. 19 GeoFide and the rise of Generative Engine Optimization A platform for getting your content cited by AI search engines. GEO is becoming a real channel, and most enterprise marketing teams have no owner for it.
  20. 20 GLM-5.2 and the open-weights step change for enterprise agents Z.ai's MIT-licensed flagship claims near-frontier agentic coding with a usable 1M-token context. The enterprise story is what this does to cost and data residency.
  21. 21 Pentesting is becoming a pipeline stage Strix, an open source AI pentest agent, is trending. The interesting question for enterprises is not the tooling, it is who approves what the agent attacks.
  22. 22 Seedance 2.5 and what native 30-second video does to media cost structures ByteDance's new video model generates 30-second 4K clips in one pass with up to 50 reference inputs. That is aimed at the most expensive line items in a media P&L.
  23. 23 Superpowers: agentic skills as an operating methodology A 245,000-star framework for giving coding agents reusable skills. The signal is that agent capability is becoming a managed artifact, like code.
  24. 24 Build Agentic Preview Environments on Your Own VPS Build isolated pull request previews on your VPS with Dokploy, PostgreSQL, dedicated workers, safe credentials, tests, and automatic cleanup.
  25. 25 How to Build a Safe Agentic Software Pipeline Build an agentic delivery pipeline where every pull request gets isolated code, data, workflows, tests, and a live Preview URL.
  26. 26 Stop Overengineering Your App From Day One A cheap VPS can handle more traffic than most teams expect. Avoid premature complexity, tune the basics, and scale only when evidence demands it.
  27. 27 Do You Really Need a CMS in 2026? Rethink CMS defaults: start with a typed ORM schema, generate admin UIs with AI, and reserve full CMS platforms for real governance needs.
  28. 28 How to Make Product Decisions Faster With Trusted Metrics Cut decision time by design: learn how analytics becomes a product, build a 10 to 30 metric spine, and set data SLAs so teams trust numbers and move fast.
  29. 29 AI Is Quietly Changing How You Should Pick a Tech Stack Tech stack choices now include an AI factor: training data, ecosystem maturity, and how reliably AI can help you ship and maintain code.
  30. 30 How I built this site... or should I say guided my agents to build it? Discover how to build a fast, database-driven site with a static shell and cache-driven fragments using Next.js, Cache Components, and AI-assisted workflows.
  31. 31 PM, Dev, QA Are Merging: Meet the Rise of the Product Engineer Uncover how AI merges PM, dev, and QA into a single product engineer. Learn to specify user needs, criteria, and metrics to accelerate delivery.
  32. 32 Stop Blaming Your Engineers: The Real Reason They're Slow Stop blaming engineers: the perfection loop slows delivery. Learn how rapid deployments, feature flags, and real-time observability speed shipping.
  33. 33 Stop Paying for Duplicate Analytics: Fix Collection First Fix inconsistent analytics by consolidating event collection with a CDP, adding governance, and restoring trust so teams can decide faster.
  34. 34 How to Ship Faster by Ditching 'Future-Proof' Over-Engineering Ship simple to learn fast: avoid premature, complex architecture for a future that may not exist. Discover how data should drive scale and momentum.
  35. 35 Tickets are comforting. Outcomes are accountable. Move from closing tickets to delivering real user value. Learn how redefining 'done' and ownership drives end-to-end impact and meaningful shipping.