motyl.dev Weekly #16: Week 24 (Jun 8 – Jun 14, 2026)

A curated digest of what I found worth reading this week.

This week the conversation kept circling back to a single tension: AI accelerates the typing, but it does nothing for the thinking. From agentic code review to the architecture behind AI data centers, from model evaluations to contract-style comments that keep agents honest, the standout pieces all argue that human judgment is becoming more valuable, not less. Here's what I found worth your time.

🤖 AI

AI is code, and can't be prompted into being smarter A reminder that models are deterministic software, not magic, and that better prompts have limits. Real capability gains come from the system around the model, not incantations.

What is the A.G.E.N.T.I.C. Framework? A structured mnemonic for thinking about agentic system design, breaking the moving parts into something you can actually reason about and explain to a team.

The State of AI: Post-Training Agents A survey of where post-training techniques and agent architectures stand right now, and what's driving the next wave of capability gains.

GLM-5.2 vs Kimi K2.7 Code: Which Model Is Better at Planning vs Building? A head-to-head of two open models on the split between planning and execution, useful if you're picking a model for an agentic coding pipeline.

💻 Coding

Agentic Code Review How AI agents are reshaping the code review process, from triaging diffs to flagging subtle regressions, and where a human reviewer still needs to stay in the loop.

Contract-Style Comments (CSC) for the Agentic Epoch A proposal for writing comments as explicit contracts, intent, invariants, and constraints, so AI agents can edit code without breaking the assumptions a human held in their head.

Claude Code Works Better When You Let Sessions Die Counterintuitive but practical: long-running agent sessions accumulate context rot. Starting fresh often yields better results than fighting a degraded conversation.

🎨 Frontend

Creating Memorable Web Experiences: A Modern CSS Toolkit A tour of the modern CSS features, scroll-driven animations, view transitions, and more, that let you build distinctive experiences without reaching for JavaScript.

🛠️ Tools

Introducing the MDN MCP server MDN now ships an MCP server, letting your AI tools query authoritative web platform docs directly. A small but meaningful step toward agents that cite real references instead of hallucinating APIs.

Introducing eve Vercel unveils eve, its latest addition to the AI-native developer tooling stack. Worth a look if you're building on the Vercel platform.

🏗️ Architecture

The Secret Architecture Behind AI Data Centers A clear breakdown of what actually goes into AI infrastructure, the networking, power, and cooling that make large-scale training and inference possible.

📺 Watch

Kent C. Dodds interviews Grady Booch on software architecture, human judgment, and AI's limits This one's a treat: Kent C. Dodds sits down with Grady Booch, and it's not the easy, agreeable conversation you might expect. Grady pushes back hard, repeatedly landing on a different view than Kent, and the friction is exactly what makes it worth your time. The result is a genuinely interesting hour on where AI helps software design and where human judgment stays irreplaceable.

Curated by Grzegorz Motyl. Subscribe for weekly updates.