The 5,000-Year Wait Is Over: Writing is Starting Evolve for the First Time—Features My Personal Model Writing Stack

Let me set the table here with a few facts. I promise you this all connects in so stay with me.

1. Claude burns ~15x more tokens in an multi-agent configuration than it does in chat.

2. This is because Claude is doing multiple language manipulation tasks at once.

3. Writing is ~5200 years old, and it hasn’t changed much in that time.

4. Mechanical code is 220 years old, and serious code is much younger (~70 years).

5. Coding has evolved more since 1960 than writing has evolved since 3200 BC.

6. Coding has evolved more because coding is compute congruent!

7. It was built for computers, it evolved with them.

8. So it’s not surprising that how we code exploded as compute exploded—something like 200x multiple on deployment speed at scale now vs. a few decades ago.

9. Writing did none of these things. Writing was just bolted on to computers.

10. Heck, it even looks like paper on a screen.

11. That’s because writing is 8-16x more complex than code.

12. Computers lacked the compute to process language correctly for decades.

13. So we haven’t even had the option to get this right until … about now.

And now it’s all about to change. This article slash podcast is packed. It includes: reflections about where writing is going, my personal workflow on writing, why I’m not just using ChatGPT, why I’m paying for Claude, a third tool I’m using too—my current stack is 4 models deep actually—plus a brief manifesto on where writing is going, and a framework to think beyond the model about writing so we get less stressed about particular model choices! There’s a ton here so I hope you enjoy.