chrisweekly

Oof. YAGNI. 150k tokens is where you start hitting the "dumb zone" (model attention issues and inconsistent adherence to instructions).

FWIW I'm also on macos 15.7.9, zsh 5.9, oh-my-zsh - and I do get this behavior (TIL).

type eg "rg" then up or down arrow, and the history shown is filtered by that rg prefix.

Yeah, I'm well aware. I shpild've been clearer I was suggesting a 1-line entry like that in CLAUDE.json, which can pay dividends in keeping context lean - which in turn is essential for avoiding the "dumb zone" (over 125k-150k tokens), precisely the effect you describe. Experts like mattpocock suggest keeping context as lean as possible for this reason.

The "dumb zone" threshold is fuzzy but comes waay before 250k tokens. Like half that.

Ken Follett (of "Pillars of the Earth" fame) wrote a terrific semi-historical* novel called "Circle of Days" that portrays a vivid picture of the people and circumstances that could've led to the creation of Stonehenge.

*"semi" qualifier bc there's so little known for sure -- but it's at least plausible, and his extensive research doesn't contradict things that can be verified.

"Avoid use of "mannered" speech in your responses."

goes a long way for Opus and Fable.

Yeah, a couple years ago (sorry, no citation handy) I wasn't too surprised to learn that a Windows desktop can run faster in a VM on Intel Mac hardware than directly on a similarly-priced PC.

I read it as sarcastic, illustrating that it's not so trivial. (shrug)

I appreciate how Graham set himself a high bar for what success looks like for his exploration of AI-assisted coding.

A takeaway for me (drawn from what he wrote^1) is that a big increase in velocity can unlock the ability to incorporate things learned along the way into a JIT / as-you-go full redesign (for coherence) vs traditional approach (slower, incremental, features added closer to release can't influence overall design).

Also I'm really impressed by Graham's approach; treating all 3 of (monkey-patching, unit testing, observability / instrumenting / tracing live running apps) as facets of the same whole... it's just beautiful.

1. https://grahamdumpleton.me/posts/2026/08/introducing-wraptur...

Awesome! Thanks for distilling the (significant) value prop and highlighting the author's bona fides, Simon!

It's signal in the noise (which noise includes both the annoying/distracting "mannered" claudisms in the post, and the inevitable complaints in the comments -- which nearly drowned the signal).

EDIT: Also thanks for the link to his index post! Graham's expertise and generosity and suitability as a library maintainer are evident, e.g.:

  > Learning it by doing   
  
  > Reading about a library only gets you so far, so alongside the posts there is now a wrapture-workshops repository on GitHub containing 24 workshops you can work through to learn wrapture in an interactive workshop format. Each takes one thing you might want to do with wrapture and walks you through doing it in a live JupyterLab session, with the instructions in a side panel whose actions drive the session and check your work as you go. The early workshops track the blog posts, one per post, and the later ones go into areas the posts have not covered yet, such as using wrapture with pytest properly, converting an existing mock based test suite, async code, patching third party libraries, distributed tracing across two processes, and writing an instrumentation package of your own.  

  > You need nothing installed to try them.

Sorry, TLDR. Seriously, this is just not how to communicate:

> "To tune your perception make sure to give yourself enough headspace to allow the feeling pre-drinking to temporarily burn into memory ready for later retrieval when you'll reflect back on your hydration actions of the past 2~5 hours. Then you either drink or not drink (usually you'll soft-force by picking whether to go up and fetch a full cup or do nothing (I.e., just keep doing whatever you're engrossed in right then) until next time you remember the idea of hydrating)."

My 13pro is at 84% capacity, but in the last month or two it seems to lose charge much more rapidly than it used to. Probably the only reason I'll consider an upgrade.

I've often read (sorry, no citations handy) that by the time you notice / feel thirst, you've already been dehydrated for some time. And chronic (mild) dehydration is much more common than hyponatremia. I'll always err on the side of drinking water whenever it occurs to me (and deliberately starting my morning with 16-20oz on waking) instead of waiting to feel thirsty.

TIMTOWTDI. "[T]hings shared with the team (in the repo)" implies project-specific skills. See e.g. https://github.com/mattpocock/skills for a stellar example of third-party skills for battle-hardened workflows.

Just use a proxy and log everything.

> "Imagine being a 1950s “human calculator” on the dawn of a computer revolution. That’s what it feels like right now."

My personal / family history is a real-world example of that evolution. My grandfather was a "computer", my father was a traditional "programmer" (lots of Perl), and I'm a SWE / frontend architect / budding "AI Engineer".

Hmm. Examples of horrible alignmnent don't necessarily outweigh the fact that most people, most of the time, mostly behave in a way that is socially aligned. (Though I'm speaking in terms of intent, conveniently ignoring the side effects / negative externalities of our collective behavior.)

Tangent: Anne-Laure Le Cunff is a neuroscientist with a talent for community-building and writing, and bringing evidence-based approaches to bear on practical solutions to various domains. See eg https://nesslabs.com

(FYIGM = F* You, I Got Mine)

Try smolvm microvms from https://smolmachines.com - among other benefits they only consume host resources if they're actually used.

See also the amazing https://betterexplained.com -- one of the best pedagogical resources I've ever encountered.

Don't feed the trolls

Cool! S an avid reader, Whispersync is one of the things that keeps me paying Amazon/Audible. Competition and viable OSS alternatives would be great!

"completely" LLM generated?

Hard boundaries there simply don't exist for most workflows, and I suspect if they did, it'd be at the expense of things ~nobody would want to give up.

Simple enough to script someth to do just that.

Great approach - I'd add a key one: "Who else should I talk to / who is (regardless of title) effectively responsible for X?"

Zed still has a lot of rough edges. A month or two ago I couldn't believe that a basic feature like code folding was just completely broken.

Hardware requirements?

I don't like htmx.

Compared to component-based approaches, its template / DOM / HTTP-fragment contracts are harder to share, type-check, version, and compose.

Application behavior is distributed among:

  - Markup attributes and DOM structure  

  - HTTP endpoints that return fragments rather than whole documents  

  - Swap rules and event hooks  
  
  - Form / query-string / hidden-field state  

  - Any supplemental JavaScript for widgets and client-only behavior  

IME, htmx doesn't remove complexity, it just relocates it.

Context: doing webdev for a living since 1998.

I highly recommend the book "Deep Simplicity" by John Gribbin.^1

It explores precisely this topic: the emergence of unexpectedly rich, complex behaviors from very simple rules given feedback and sensitivity to initial conditions.

1. https://www.penguinrandomhouse.com/books/71851/deep-simplici...