A shorter batch this week. Here are the three new recommendations.
Your Agent Framework Is Just a Bad Clone of Elixir — George Guimarães
This is the most interesting technical argument I’ve read in months. The thesis: the actor model that Erlang introduced in 1986 is the agent model that AI is rediscovering in 2026. Every pattern the Python AI ecosystem is building — isolated state, message passing, supervision hierarchies, fault recovery, already exists in the BEAM virtual machine. And it’s been running telecom switches, WhatsApp, and Discord at scale for decades. The comparison table alone is worth the read: problems that agent frameworks like LangGraph and AutoGen are still solving today map almost exactly to primitives OTP had formalised by 1998. He also makes a compelling practical case for why Python can’t fully close the gap, the remaining 30% (preemptive scheduling, per-process garbage collection, hot code swapping, true fault isolation between agents) requires runtime-level support that you can’t bolt onto a language that wasn’t designed for them. Note: there’s a follow-up post where he addresses the fair criticism this received on Hacker News, which is equally worth reading.
How to Do AI Analysis You Can Actually Trust — Caitlin Sullivan (Lenny’s Newsletter)
This is a guest post on Lenny’s newsletter by a UX research veteran who has spent over 2,000 hours testing customer discovery workflows with AI, and the depth shows. The core problem she identifies is that AI output always looks confident regardless of accuracy, these mistakes are invisible until a stakeholder asks a question you can’t answer, or a decision falls apart three months later, or you realize the “customer evidence” behind a major investment actually had enormous holes. The model comparison alone is useful: Claude is best for thorough analysis with depth and nuance; Gemini delivers fewer themes but with stronger grounding; ChatGPT excels at packaging findings for stakeholders but is the most prone to combining quotes into Frankenstein statements that never actually appeared in your source data. The quote verification prompt snippet is something I’m adding to my own workflow immediately. The full article is behind Lenny’s paywall, but the free portion covers the most actionable material.
How I Use Obsidian — Steph Ango
Steph is the CEO of Obsidian, so this is as authoritative as a vault setup guide gets. What I appreciate here is how minimal and principled his system is, no nested folders, no complex taxonomy, mostly everything living in the root of the vault. His system is oriented towards speed and laziness. Rather than deciding where things go, his notes are primarily organised using a “categories” property, and he navigates almost entirely through the quick switcher, backlinks, and links within notes. The fractal journaling approach is worth borrowing: individual thoughts captured throughout the day with a unique note hotkey, compiled every few days, reviewed monthly, reviewed yearly. The result is a fractal web that can be zoomed in and out of at varying degrees of detail, letting him trace back where individual thoughts came from and how they bubbled up into bigger themes. He also shares the vault template as a free download on GitHub, which is a good starting point if you want to try his setup directly.