Tomas Marek

23 posts

Tomas Marek banner
Tomas Marek

Tomas Marek

@tomasmarekk

Researching LLMs and agentic workflows | AI Driven Development | also official @Webflow Partner & ex-FE developer.

The Simulation Katılım Ekim 2018
281 Takip Edilen104 Takipçiler
Tomas Marek
Tomas Marek@tomasmarekk·
I genuinely don’t get Anthropic’s endgame here. From their POV it is not ideal that people use the sub inside another harness, but these people still pay for the sub anyway. This vendor lock-in gets you less dolla from subs and more bad PR on social media. And if they seriously think they can build their own harness and outcompete open source stuff like OpenClaw, that is just delusional. I would replace that whole sunshine marketing dep with people who think rationally, and they would be back to tier 1 AI lab within a month. I really do not get it.
English
1
0
1
215
kitze 🛠️ tinkerer.club
kitze 🛠️ tinkerer.club@thekitze·
banthropic now blocks this if you run it anywhere claude -p --append-system-prompt 'A personal assistant running inside OpenClaw.' 'is clawd here?' a bunch of crybabies that they are lmao
English
12
1
162
8.9K
Tomas Marek
Tomas Marek@tomasmarekk·
@steipete @work_____ Whenever you see Claude worship on X, here's a six-month proven cortisol control hack. Hover over their username, read for two seconds, then move the cursor away. Instant peace. Instant clarity.
English
2
0
6
3.5K
Alejo
Alejo@work_____·
Bros openclaw without Claude is shit. I can tell you bc for 2 months I’ve tried OpenAI $200 and Claude $200 subs. OpenAI is the bitch of Claude. OpenAI is good to do some cleaning and smol cooking and that’s it. Important vibe coding is for Claude and Claude only.
Peter Steinberger 🦞@steipete

@NoahEpstein_ models auth login --provider anthropic --method cli --set-default

English
30
5
242
121.9K
Tomas Marek
Tomas Marek@tomasmarekk·
@PaulSolt Custom "github-unlobotomizer" so it stops committing files and folders that are in .gitignore 😂 That’s basically my only 5.4 complaint.
English
0
0
0
841
Paul Solt
Paul Solt@PaulSolt·
What's one Codex skill that you cannot live without?
English
40
3
127
34.2K
Tomas Marek
Tomas Marek@tomasmarekk·
@okbish @AshikArtistry sure, I agree with that. I'm a bit degenerate from the UI community. But OP was criticizing the fact that LLMs can’t produce good design (which I also agree with), but ironically the output we’re seeing here exhibits exactly those LLM-style UI patterns 🤠
English
0
0
1
53
Ashik
Ashik@AshikArtistry·
Claude Code just shipped a flawless landing page. Zero back and forth. Zero revisions. This thing did not miss. prompt below ↓
Ashik tweet media
English
48
25
629
72.3K
Tomas Marek
Tomas Marek@tomasmarekk·
@kimmonismus 👇x.com/tomasmarekk/st…
Tomas Marek@tomasmarekk

It’s honestly crazy to watch how hard people are vendor-locked into Claude Code, even after months of getting treated like crap by it. I’m not saying Opus or Sonnet are bad models. But I’m in SWE circles, and the reality is that most serious engineers have been using Codex and the GPT family since December 2025. I used Claude Code all through 2025 because for a while it really was on the edge. But that changed at the end of 2025. And the benchmarks that actually matter have been showing that pretty clearly ever since. 5.2-codex, 5.3-codex, and now 5.4-high are just a tier above Claude models on stuff like Terminal Bench and LiveBench. Even the Claude Code harness is bad enough to drag it down to #39 on Terminal Bench, while Codex is sitting at #8. Since December 2025, basically every metric has been pointing toward OAI. - better limits - smarter models - way better public communication from the devs Are there downsides? Sure. - It’s a lot slower. But it also does the work better, so over time it saves you effort. - It’s not some AI normie companion that just pats you on the head and says yes to everything. You actually have to know how to explain what you want and what you’re trying to do. So yeah, it’s just sad watching so many people stay irrationally locked into the worse option while it keeps screwing them over. Give switching a shot. It’s worth it. And I’ll be the first one to go back to Claude Code the moment it’s actually on the edge again. But it just hasn’t been since December.

QME
0
0
0
594
Tomas Marek
Tomas Marek@tomasmarekk·
It’s honestly crazy to watch how hard people are vendor-locked into Claude Code, even after months of getting treated like crap by it. I’m not saying Opus or Sonnet are bad models. But I’m in SWE circles, and the reality is that most serious engineers have been using Codex and the GPT family since December 2025. I used Claude Code all through 2025 because for a while it really was on the edge. But that changed at the end of 2025. And the benchmarks that actually matter have been showing that pretty clearly ever since. 5.2-codex, 5.3-codex, and now 5.4-high are just a tier above Claude models on stuff like Terminal Bench and LiveBench. Even the Claude Code harness is bad enough to drag it down to #39 on Terminal Bench, while Codex is sitting at #8. Since December 2025, basically every metric has been pointing toward OAI. - better limits - smarter models - way better public communication from the devs Are there downsides? Sure. - It’s a lot slower. But it also does the work better, so over time it saves you effort. - It’s not some AI normie companion that just pats you on the head and says yes to everything. You actually have to know how to explain what you want and what you’re trying to do. So yeah, it’s just sad watching so many people stay irrationally locked into the worse option while it keeps screwing them over. Give switching a shot. It’s worth it. And I’ll be the first one to go back to Claude Code the moment it’s actually on the edge again. But it just hasn’t been since December.
Chubby♨️@kimmonismus

I don’t get this post. All users are saying they're reaching their rate limits faster than before. Anthropic investigated and said: we found nothing, but hey, why not use Sonnet instead of Opus because Opus uses about twice the bandwidth?

English
0
0
1
790
Frederik Jacques
Frederik Jacques@thenerd_be·
Today is my first day at Anthropic. Super excited I shipped my first change today, added source maps so debugging is easier. Can’t wait to show you all what I’ve been working on! cc: @AnthropicAI
Frederik Jacques tweet media
English
593
548
18.3K
786.1K
Tomas Marek
Tomas Marek@tomasmarekk·
@trq212 I swear someone in your marketing team has been sabotaging Anthropic for the past 5 months, because I genuinely can’t explain your moves any other way. There’ve been like 50 already where I just shook my head 🤦
English
0
0
1
424
Thariq
Thariq@trq212·
To manage growing demand for Claude we're adjusting our 5 hour session limits for free/Pro/Max subs during peak hours. Your weekly limits remain unchanged. During weekdays between 5am–11am PT / 1pm–7pm GMT, you'll move through your 5-hour session limits faster than before.
English
2.3K
527
7.4K
7.6M
Tomas Marek
Tomas Marek@tomasmarekk·
Please don’t. I understand why it might help from a marketing perspective, but personally I treat every GH repo with “co-authored by Claude” as irrelevant and immediately put it in the “rushed, unserious project” bucket. It’s already easy enough to spot shitty GH projects, and this would just add even more pointless noise (at least in my delusional world view 😅).
English
0
0
1
231
Tibo
Tibo@thsottiaux·
Do people like this? We don't do this for codex because it exists to help you and it's important that you remain the owner and accountable for your work without AI taking credit. At the same time it does mean that you can't trace how popular codex is among repos.
Yuchen Jin@Yuchenj_UW

I noticed something interesting: Claude Code auto-adds itself as a co-author on every git commit. Codex doesn’t. That’s why you see Claude everywhere on GitHub, but not Codex. I wonder why OpenAI is not doing that. Feels like an obvious branding strategy OpenAI is skipping.

English
595
75
2.6K
369.9K
Tomas Marek
Tomas Marek@tomasmarekk·
@MercuryBarbora Jako jediný problém tu vidím zanedbanou docházku na vzdělávací platformu dodiri .cz - za 5s tenhle pes zabije člověka a senil jehož největší challenge je ustát svůj další krok s tím fakt z 60m nic neudělá.
Čeština
0
0
2
85
Barbora Mercury 🇨🇿✌️🇪🇺
Barbora Mercury 🇨🇿✌️🇪🇺@MercuryBarbora·
Tady z téhle naprosto zbytečně vyhrocené situace v lese mi vyšli jako nejchytřejší ti dva psi. A přijde to jen mně, že ta ženská je docela hysterka?
Čeština
191
5
149
27.2K
Tomas Marek
Tomas Marek@tomasmarekk·
@Yahiko1239170 @TukiFromKL the only thing I’m thinking about is where you’re getting these numbers from. Does OAI publicly share how many users left? Highly unlikely, so is this just a subjective figure based on an X narrative? :D
English
1
0
0
91
Yahiko
Yahiko@Yahiko1239170·
@tomasmarekk @TukiFromKL ~95% are free subscribers, around ~5% were paying subscribers. From these, people are leaving: ~700k-800K which were 0.1% paid subscribers left between 29 January and 13 February 2026. and now The 1.5M figure also comes from those paying subscribers, so they left too. Now think.
English
1
1
6
273
barneytheboi
barneytheboi@barneytheboi·
Is anyone else’s ai bot acting weird?
barneytheboi tweet media
English
306
324
13.7K
867.5K
Tomas Marek
Tomas Marek@tomasmarekk·
I get the impression and rage bait farming, but this just harms users. So it needs to be said that what you’re writing is complete nonsense. Every senior developer is using Codex right now. The Pro plan is basically inexhaustible (almost), plus you get additional features in the web app. Sloptopus at $100/month lasts about three days of heavy work before you hit the limit. You’re all delusional, just open up Codex supporters on X and it’s always a senior. Then open up Slopus supporters and you always see some Fortnite random 😂
English
4
1
71
9.8K
BURKOV
BURKOV@burkov·
Codex on the $20/month Plus plan lasts for ~1.5 days of work and then blocks you for 3 days. And you have to be a complete idiot to pay for the $200/month Pro plan for Codex instead of the $100/month Max plan for Claude Opus.
English
187
19
864
191.8K
Tomas Marek
Tomas Marek@tomasmarekk·
@naval Slop code was always the baseline, 98% slop, 2% great, throughout internet history. AI doesn’t change the ratio, it just multiplies output, so you’ll see 100x more of both.
English
0
0
2
43
Naval
Naval@naval·
We’re now in the era of slop code.
English
1.4K
575
11.1K
1.5M
Tomas Marek
Tomas Marek@tomasmarekk·
Interesting to watch how the narrative among senior devs has shifted since the release of GPT 5.2-codex-xhigh and Opus 4.5. That switch happened within a 1‑month timeframe, which is insane 😅. But I don’t think strong senior devs should be worried. It’s just another abstraction layer (Binary, Assembly, Procedural (Fortran, C,..), OOP (C++, Java,..), Declarative, and now LLMs). Senior dev knowledge isn’t going anywhere, and they’ll still be the ones pushing the edge. There’s basically no chance a “pure vibe” person without fundamentals of the lower abstraction layers will do better work than someone who has them. That’s my prediction for 2026, and it’ll hold until “AGI” arrives. No reason for depression, more like being happy about the competitive advantage.
Duca@big_duca

I am not sure if other developers feel like this. But I feel kinda depressed. Like everyone else, I have been using Claude code (for a while, it’s not a recent thing lol). And it’s incredible. I have never found coding more fun. The stuff you can do and the speed you can do it at now. Is absolutely insane. And I’m using it to ship a lot. And solve customer problems faster. So all around it’s a win. But at the same time. The skill I spent 10,000s of hours getting good at. Programming. The thing I spent most of my life getting good at. Is becoming a full commodity extremely quickly. As much fun as it is. And as much as I like using the tools. There’s something disheartening about the thing you spent most of your life getting good at. Now being mostly useless.

English
0
0
2
697
Tomas Marek
Tomas Marek@tomasmarekk·
@big_duca been there, it’s normal. LLM dev is just another abstraction layer, you’ll reuse basically everything you learned before. You’ll still have an edge over pure “vibe-coders” who don’t understand software.
English
0
0
1
47
Duca
Duca@big_duca·
I am not sure if other developers feel like this. But I feel kinda depressed. Like everyone else, I have been using Claude code (for a while, it’s not a recent thing lol). And it’s incredible. I have never found coding more fun. The stuff you can do and the speed you can do it at now. Is absolutely insane. And I’m using it to ship a lot. And solve customer problems faster. So all around it’s a win. But at the same time. The skill I spent 10,000s of hours getting good at. Programming. The thing I spent most of my life getting good at. Is becoming a full commodity extremely quickly. As much fun as it is. And as much as I like using the tools. There’s something disheartening about the thing you spent most of your life getting good at. Now being mostly useless.
English
1.2K
421
7.5K
1.2M
Tomas Marek
Tomas Marek@tomasmarekk·
@nateberkopec actually first AI-generated UI I’ve seen that doesn’t scream “AI.” I’m deep into AI FE capabilities (native ones, figma MCP etc.), but this one lacks the usual vibe. Why do you think that is? It can’t just be dark mode, everyone has that now.
English
2
0
3
1.9K

Keşfet

@steipete @work_____ @PaulSolt @okbish @AshikArtistry @kimmonismus @thenerd_be @AnthropicAI