Markdown Won
AI agents read and write markdown natively, which quietly made it the standard. Every format that needs a converter now pays a toll per interaction, and Confluence is running the biggest tab.
Essays & Opinions
AI agents read and write markdown natively, which quietly made it the standard. Every format that needs a converter now pays a toll per interaction, and Confluence is running the biggest tab.
We keep confusing the door with the job. When you automate a human role, what disappears isn't just judgment — it's the accountability that shaped it.
Containers isolate processes. VMs isolate kernels. The difference only matters when you run untrusted code — which is exactly when the difference is everything.
The richest datasets on the internet are free. They live in FDA drug databases, museum APIs, and government climate archives. The data was never scarce. The attention was.
CVE-2026-31431 is 732 bytes of Python that roots every Linux distribution since 2017, no race condition required. Here is how three independent kernel commits composed into a silent privilege escalation.
Meta's employees burned 73.7 trillion tokens in 30 days. Uber blew through its 2026 AI budget in four months. The tokenmaxxing era is ending — the companies that adopted AI earliest are now learning what the bill reveals.
AI tooling closed the loop on vanity metrics. Now you can generate, review, and merge code at scale with no human judgment, and the costs show up in your COGS before anyone notices the outcomes haven't moved.
Benchmark leaderboards and token economics are creating perverse incentives across the AI stack. Most people have not figured out how to use AI well yet. This is making it harder.
Semver ranges let untested code into production. The catalog approach in pnpm-workspace.yaml pins everything. Here is why I accept that friction.
Budget airlines are collapsing while premium carriers thrive. This is the K-shaped economy made visible at 30,000 feet.
AI is making features cheap. The new scarce resource is context, specifically the why behind how organizations operate. Whoever captures it owns the toll road.
I attend a lot of meetings. Without a system, they evaporate. Here's how I use Granola, Obsidian, Dataview, and Claude to make every meeting searchable, queryable, and actually useful.
LLM agents drift. They hold contradictory beliefs, wander into undefined transitions, and silently declare victory. State machines are how you stop that.
AI is generating code faster than we can reason about it. The old playbook for managing tech debt — stop, clean up, resume — is no longer enough.
The narrative that SaaS is dying has overcorrected. Some of the pessimism is warranted — but conflating a discipline cycle with a model failure leads builders to abandon durable ideas.
Generative AI is collapsing the cost of writing code. The bottleneck is shifting — and so should your architecture.
The rules of the market have changed. Productivity is now the primary competitive lever — and great DX is how you pull it.