Claude-pilled, but no longer on Claude
Will H26 Aug 2026Founding Agentic AthleteDagny
2 min read

If you're anything like me, which if you're reading this means you undoubtedly are, the moment first having Claude Code felt like magic.
You opened this strange terminal with a matrix like feel and started to chat.
And then blocks of text turned into the first strands of example html links that started to demonstrate what was now possible.
If you’re also like me, you started with a personal pet project like a CRM that you’ve always wanted to build.
And then came learning about GitHub, building skills, and the slow drip wait until your token limits were refreshed.
You may have even tried your hand with mutli-agent personas through a paperclip or multica.
You were claude-pilled.
Flash forward, and you’re found yourself on a roller coaster of highs and lows.
You’ve figured out how to package up specific workflows into a set of commands and have multiple repos that are brimming with context files to speed up the learning process.
You also come to realize that your personal CRM project is not actually that useful, and you prefer to modify existing SaaS tools or if you’re adventurous graft something from open source.
And then there’s Claude.
Cowork came out, and you were already used to the terminal and were protective of the token and context loss that a heavier front end did not abide by.
And then there was Fable.
Ephemeral, magical once again until the restrictions took over and started what is now the quagmire that is taking hold.
With the re-introduction of Fable came the new class of Opus (5.0) models.
And this is where the tide shifted.
The more I used these new classes of Opus, the more fatigued I began to felt.
While this new class of model was meant to be more thoughtful and progressive, the verbosity and ‘opinionated’ viewpoints felt like I decided to reverse time and take the blue pill.
Even caveman prompts could not protect us.
Simultaneously, new models came out from OpenAI, Gemini, to Grok which were starting to capitalize on improvements, usability, and cost.
Grok bot especially is a mirror into how what we considered the most important tenants of building in the phase 0 is becoming obfuscated and will continue to do so.
And while I still proudly consider myself claude-pilled as I will no longer think and work the same, it’s a stark reminder that we are still in the earliest innings of what making these tools more accessible.
Read next
- How 'Shadow' agents tuned my Grok Bots ↗
- The paradox of time ↗
- The compounding effects of high-quality data infrastructure ↗
- A specialist can survive ↗
- It's hard to teach AI ↗
- Don’t be afraid to make a mess and learn ↗
- There is no such thing as AI mastery ↗
- Why multiplayer AI has been out of reach ↗
- Still stuck in spreadsheets? you’re not alone ↗