🏢 Developer @JetBrains 🍃 vegan 🐈 crazy cat person
- Tweets 1,651
- Following 307
- Followers 183
- Likes 2,416
5.6 Sol was my daily driver. There was no task it couldn't solve at reasonable budget usage. Then they nerved it and eventually gave us 6 Sol which doesn't compete at all. Somehow now a 100$ OpenAI sub feels less usable than a 20$ Claude sub.
1
35
Andreas Eisele retweeted
NeoVim + IntelliJ IDEA backend for Java & Kotlin.
Very experimental and rough around the edges, but already a lot of fun! 🔥
Who wants to give it a try?
@ThePrimeagen @teej_dv @sudobunni @Folke
@nicknisi @joshmedeski @Exosyphon
@elijahmanor @dillon_mulroy @lewis6991
@echasnovski @justinmk @Neovim
5
17
32
3,819
Andreas Eisele retweeted
Today we introduce JetBrains Air - the product system for software development orgs in the age of agents.
Built for developers, team leads & engineering directors.
Open and flexible: steer and control almost any LLM & harness.
Read more:
jb.gg/air-announce
79
78
464
247,609
Not sure if it was a bug but recently codex started to display the internal model monologue instead of its "working..." banner for me. It's rather eye-opening what those text generation models go through to simulate intelligence.
2
43
I wonder how much or idea of model capability is just "a feeling" in our brains. GPT Sol let me down yesterday with some bad outputs and now I suddenly "feel" that Opus 5 would be way smarter.
46
Huh, codex CLI now has "queued questions"? Never noticed this before. Rather neat that the agent can stay busy and at the same time do a "here's a related question for you" prompt.
40
What did they do to my boy "gpt 5.6 sol"? It now also started to feed basic operations through multi line python scripts. That's not how sol acted for the last weeks.
51
One of my hobbies for the past months was implementing song ideas via @suno. The v5.5 model was quite good but Sony and Gema et. al. had issues with the training data. So now Suno tells us we have to use the v6 model. It might be good but it doesn't hold up to v5.5's quality.
1
1
117
Astra was eating my weekly usage like crazy. So I thought to use delegation to sol, terra and luna. Everything sounded nice on paper. The coordination overhead was even worse than plain astra. Nice model you got there, I just can't use it at all.
1
47
Strange situation for coding agents currently. Claude is too fond of adhoc python helpers for simple file edits across Opus 5 - Fable 5.1 while GPT Astra apparently lost all sense for maintainability. Not that generated code was super maintainable in the first place.
1
96
what’s the German compound word for the feeling when you listen to an album on Spotify and still expect the exact tape glitch that was on your home-recorded cassette 30 years ago?
1
46
Andreas Eisele retweeted
Code review isn’t just about finding bugs. It’s how teams share knowledge, teach junior engineers, build collective ownership and spread architectural understanding.
Rachel's Ramblings:
Maybe We Shouldn't Be Reviewing All This Code: Or, perhaps the problem isn't that AI has broken code review, maybe it’s that we've been using code review to solve the wrong problems
martinfowler.com/rachels-ram…
1
4
150
I wonder when the first network operators will start selling "AI grade" internet access. Same as when they traffic shaped streaming into higher cost tiers.
36
Random LLM superpower: security analysis of unknown binaries. Just tell your agent to disassemble and detect any malicious or problematic calls. Don't trust it though. Also do it in a sandbox / VM of course.
34
Andreas Eisele retweeted
Looking forward to hosting #IntelliJIDEA Conf again this year.
I will be hosting talks by @holly_cummins @mthmulders and @antonarhipov, other speakers include @starbuxman @JamesWard @ammbra1508 @thomaswue @shelajev @jbaruch @gAmUssA and more.
Join us for this free virtual event!
One week to go!
IntelliJ IDEA Conf 2026 starts soon – a free virtual conference for JVM developers.
🗓 September 8–9
👉 Register now: jb.gg/5vx31c
#IntelliJIDEAConf
1
7
24
5,170
