I'm so tired of ghost writers, so tired of post templates, so tired of people "comment X to receive Y"
I blocked so many people. LinkedIn actually makes it so cumbersome to block someone, you have to actually go on the op profile page to do it
37
Alas once again the quality of the report generated by Opus beat Sol
Sometimes I still need my agents to generate pdf reports (mostly data stuff). They're useful but they're usually way too verbose.
I simply asked my agent to generate a pdf like a slide deck instead, the result is SO GOOD.
Proper visuals when needed, and so NOT verbose.
43
Sometimes I still need my agents to generate pdf reports (mostly data stuff). They're useful but they're usually way too verbose.
I simply asked my agent to generate a pdf like a slide deck instead, the result is SO GOOD.
Proper visuals when needed, and so NOT verbose.
Everybody saw the fuss around the @OpenAI's 6.2 model so I gave them a real shot today.
Astra 6.2 for research, "creative" questioning, ideating, "smart thinking" etc, is overall good. What it output was helpful and the prose was overall okay.
I did not tested it on coding so I have no opinion on that. Neither did I tested it on 3D stuff, but I've only heard good things there.
I however did test Sol 6.2 on long running tasks (working on a migration + refacto atm) and hit some of the weird issue others reported:
- It paused a sub agent multiple times waiting for my approval to run some basic python scripts sol 6.1 usually run without any issue.
- It paused one to... ask whether it was okay for it to use "worktree slot 6 instead of 5" (I keep a worktree pool because creating worktree on windows is slow) because it was not sure whether the slot was available. Same here, other models never had any such issue. It's definitely not the kind of blocking task I expect to be pulled into.
- As sub-agents, Sol 6.1 shipped about as much as 6.2 for ~4x fewer tokens, and 6.2 burned 74% of my weekly Pro quota in 5.5 hours (my bad for testing ultrafast I guess though)
After a day burning tokens on sol 6.1 and 6.2 today (and opus 5.5, and muse wonder full), I guess my conclusion would be:
Astra 6.2 is fine for very specific tasks, but atm I share the same feeling as others and 6.2 is not up to the expectations.
Unless it improves significantly in the coming weeks, I unfortunately don't think I'll keep my 200 sub as I get more done for the same money with Opus. Also as a solo freelancer I cannot justify stacking 200 subs across providers without really feeling the difference.
It's sad, because I really like switching between sol and opus, but sol doesn't feel worth it right now.
90
You can even find discount in the page 👀
“Shopping is fun” is our mantra. Now, shopping on @Shopify's own merch store is 600% more fun. While you're there, cop our brand-new BFCM calendar, lovingly designed to help businesses prepare for the biggest weekend of the year … all year.
Link in thread 🧵
78
Aurélien retweeted
Announcing Oxlint + tsrs
A drop-in replacement for tsgolint for type-aware linting & typechecking.
Up to 7× faster, with up to 75% less memory.
oxc.rs/blog/2026-11-17-faste…
Really enjoying that the @AnthropicAI banked reset didn't push back my regular reset date. Unlike @OpenAI, where using it moves your next reset to 7 days after you activate it.
77
Aurélien retweeted
Introducing Fluid Functionalism Foldkit
A port of @micka_design's Fluid Functionalism to @foldkit & Foldkit UI
The goal was to be faithful to the source without reimplementing the minutia of the underlying component libraries
GitHub: github.com/timhanlon/fluid-f…
Demo: fff.timhanlon.com/
Likely some rough edges, and the newer components like data table are still WIP
Nice, seems like @OpenAI is aligning with @AnthropicAI and now offers 200$ credits as well with its pro plan
85
This reminds me of a funny story.
A couple of years ago, I was freelancing as a software dev for a company running a B2B ecom platform. My daily rate was below market for an equivalent role, but I'd accepted it because the company seemed "cool".
They had an important month coming up, with a lot of showrooms and events, so we had to work a lot. Like... a lot. Sometimes even on weekends.
I remember one Tuesday, the daily call started with an urgent feature we HAD to ship by the next day. Another dev and I worked on it from 9am to 1:30am, almost non-stop. Only to learn the next morning at 8am that the so-called urgent feature had been dropped for something entirely different.
It s*cked, but I guess 1) it's business, it happens, and 2) I should have drawn the line myself when the squad manager couldn't do it for us.
In the end, we delivered every requested feature on time, fixed every critical issue within minutes/hours whenever there was an emergency, and on the software side, the events were overall a success.
At the end of the month, I sent my invoice, which was obviously higher than usual given all the overtime.
That same day, the CEO (not the manager, the goddamn CEO) asked me to hop on a call. He was absolutely furious, even though our squad manager had explicitly asked us to work this overtime (nothing was done behind anyone's back).
Instead of thanking us for being available way outside our usual schedule, he scolded us like children, in the most condescending way, for charging more than usual.
The minute I left the call, I opened a Google Doc, wrote my resignation letter, and quit the same day.
3
5
504
I really like it. I think for people used to play with ai chats it's great. But I wonder whether it would really be helpful for "normies"
AI interfaces nowadays are still pretty much turn-based.
In my PhD thesis, I thought about what interaction would look like if we could go beyond taking turns.
One example is intra-turn interaction:
1
76
A prompt:
Audit all local AI-agent logs for complete secrets exposed in model context. Report EVERY finding: project, key, agent/model, date, evidence + CSV/HTML.
Validate locally; exclude false positives. Mask before model input (prefix…last 4). Never read current .env files.
64
Ran a quick audit, turns out we have a few keys we need to rotate
Probably going to ban .env files in my projects and for the people I work with, even for dev keys.
LLMs can be so clumsy, they sometimes read the content and add it to the chat context. Using things like Infisical is going to become mandatory.
The fun thing is that now the agent tells you when it happens. I already got a “Oopsi I read the content of your .env.{some-env} inadvertently, you should rotate your keys hihi”. But that means it 100% used to happen before without the agent notifying us.
1
3
278
Was wondering why suddenly it felt slow to work, turns out my sub agents were spawned without Fast Mode on
63
Probably going to ban .env files in my projects and for the people I work with, even for dev keys.
LLMs can be so clumsy, they sometimes read the content and add it to the chat context. Using things like Infisical is going to become mandatory.
The fun thing is that now the agent tells you when it happens. I already got a “Oopsi I read the content of your .env.{some-env} inadvertently, you should rotate your keys hihi”. But that means it 100% used to happen before without the agent notifying us.









