GLM-5.2 is all over my feed. Everyone's saying open weights finally caught Opus.
Here's the part nobody says out loud: the people declaring open-source won still have Claude open by default. Real migration was never won by a benchmark — it's the day you catch yourself reaching for the new thing without thinking. That's when you've actually switched.
That is more of a habit change.
Show more
The part nobody's saying: the files we write for the agent — AGENTS.md, CLAUDE.md, comments — are now an injection surface.
It's highly likely that agents auto-read and execute external AGENTS.md, CLAUDE.md.
High security risk
Show more
Got em. I poison my AGENTS.md (and other things like code comments) all over the place with prompt injections like this to find people who don't review their code and sling it off to another human. Catches folks all the time and then its an instant ban.
As I've said, I don't care if you don't review your own code. But if you're submitting code to an OSS project and crossing a human boundary, it is simple courtesy to do some human review.
Show more
Most of my day is now spent being the while-loop.
Run the test, read the error, fix one line, re-run — look up and the day's gone.
So I taught the agent to run the loop itself. Paste this into Claude Code to save your time:
run the check → read the errors → apply the SMALLEST fix → re-run, until exit code 0 or pass
— with a hard attempt cap so it can't bulldoze a speculative rewrite between runs.
Show more
tok/s used to be a vanity metric for coding models. Kimi K2.7 just made it the whole game — by outrunning you.
At ~180 tok/s (and GLM-5.2 now on every Coding Plan tier), the model generates faster than you can read, let alone review.
The bottleneck quietly moved: it's not the model anymore. It's you, sitting in the loop trying to keep up.
The instinct is to slow down and check every step. Wrong move — that just reinstalls the human as the blocker.
once the model outpaces your reading, stop reading. Hand the whole loop to the agent and put a second agent on review. Your job shifts from reading lines to designing the checks. Babysitting a 180 tok/s agent is like proofreading a printer mid-print.
Show more
Given token economics, we really need @apple’s new ceo to go all in on workstations that can run local, open source models
Ideally, with a router that can flip between local models and frontier models when the former gets stuck.
And America needs an open source champion — we really should not be comfortable with the Chinese owning the open source LLM market to the extent they do
Show more
Looks nice!
Build audience
I think the challenge is that everyone can now build apps
But
1) almost nobody has distribution (like an audience), or
2) the money to pay for distribution (ads or UGC), or
3) the creative genius to get distribution for free (classically called guerilla marketing)
Show more
