micromanaging 4 AI agents looks good on a timelapse

Codex is so slow right now. What's happening

FINALLY. This is AGI.

Introducing the Fusion API, the smartest compound model in the market. Fusion achieves Fable-level intelligence at half the price. How it works 👇

So far, the fastest and most reliable UI QA workflow I've found is using Storybook. I have AI generate the controls, then verify everything in a single pass.

/fast is not so fast right now

The answer depends on the token budget you're willing to spend for each pivot. Loops or not, you still need feedback

Before AI, I had 5 unfinished projects. After AI, I have 128 unfinished projects.

I customized my linter to fail whenever AI generates "isSomething(...): value is Something" type guards. The error message simply says: "Find and reuse a Zod schema instead."

You are not tokenmaxxing if you don't have at least 20 copies of this function in your repo.

Just realized I can save a lot of expensive tokens by pushing testing, linting, and build feedback to Codex Spark subagents instead of streaming massive logs into the main context.

Starting to think that maybe Angular had it right all along…

Is it worth upgrading to Codex 20x? I use /fast so much that it feels like a compiler for thoughts. My real bottleneck isn't writing code, it's getting feedback. When feedback loops are slow, I end up batching work into much larger chunks, which makes iterations more expensive and increases the cost of being wrong. /fast lets me keep the loop tight and adjust course sooner. My biggest hesitation tho about upgrading is the pricing stability. I've switched between enough AI products to notice the pattern: you sign up for one set of limits, get used to a workflow, then a few months later the pricing or quotas change and you're effectively getting less than what you originally signed up for. It's hard to build habits around a tool when the economics keep moving underneath you.

11 minutes break time

Be aggressively kind! If you are really happy inside – share that shit. Cause the angry are better at expressing it and it’s time we change that 

I run 1-2 agents at once. Rarely more. Why? Because code generation isn’t the bottleneck. Great read.

Ratio’ing the Pope is an insane flex

This view is basically right outside our house. I grew up with a perfect volcano casually sitting in the background. Could've seen the meteor just by opening the window and touching some grass. But the usage limit just reset, and I was too busy codexmaxxing 😆

I'd rather burn through all my tokens in one uninterrupted flow session than get dragged out by 5-minute reply cycles.

It seems like Tailscale isn't working properly anymore on Starbucks wifi. I used to get a much faster connection when routing through my home PC as an exit node.

Back to walking again after being sick the past week. Slowly getting my energy back.