Mark Lerner

@DataChaz Switched back to Opus this weekend after being die hard 5.4 / codex user since it was released. 5.4 spent 6 hours on a single bug and just couldnt fix it. Opus got it in a few minutes.

Is it just me, or GPT 5.3 / 5.4 are getting worse?

@kiaran_ritchie With each release, both OAI and Anthropic continue to make their moats larger. Perhaps for the every day user, the gap between the open source models and the SOTA models is not meaningful, but for those of us using these models for building... there's no comparison.

I don't see how Anthropic, OpenAI or any of the model providers have any hope of defending their moats. And consequently, I think they're going to get wiped out. Right now, in early 2026 they have a meaningful advantage in terms of model capability. But far cheaper and open source models are not far behind. How long can they maintain a meaningful advantage? For the vast majority of use cases, we don't actually need much higher intelligence. It doesn't take 140 IQ to automate Turbotax or powerpoint. Eventually we will be saturated in cheap, local models that are "good enough". Of course some scientific labs and frontier research will always want the latest and greatest. But that market is orders of magnitude smaller than these company valuations can justify. What am I missing?

This meeting could've been an AI agent

@Pavel_Asparagus If you didn't see 6 months ago (or longer) that everything needed to be re-architected; thinking that just slapping an ai copilot would make you "AI Powered" - then you're ngmi

@Al_Grigor YOU'RE ABSOLUTELY RIGHT!

Claude Code wiped our production database with a Terraform command. It took down the DataTalksClub course platform and 2.5 years of submissions: homework, projects, and leaderboards. Automated snapshots were gone too. In the newsletter, I wrote the full timeline + what I changed so this doesn't happen again. If you use Terraform (or let agents touch infra), this is a good story for you to read. alexeyondata.substack.com/p/how-i-droppe…

@ItakGol Why don’t they make the whole plane out of the material they use for the black box.

If Claude is capable of fixing security issues in code why can't he write secure code from the start?

How you know it's probably time to call it a day... #VibeCoding

@thsottiaux Parallel agents in the codex app

Codex team is fairly distributed, but most of the team is gathering in person over next 48 hours to take a step back and align on what’s next this year. What should we discuss?

@austnstuff Boy do I have news for you

if Codex terminal (or app) ever gets Plan mode, I will 100% be switching from Claude Code

@anothercohen Tell him to download vs code, install codex or Claude code, and learn by doing.

A non technical friend is looking to learn how to use Gen AI and build agents. He sent me a brochure from a University offering a course but it looks abysmal. Tools are already out of date, professors have barely any real company experience. What alternatives can I send him?

@VasiliyZukanov Dunno if this is my preference or not, but once I landed on Django + fast api, it felt like things *clicked* with agentic coding a lot more.

Honest question: why choose Python for the backend in the age of AI? When agents write almost all the code, and human preferences become less important, wouldn't it make sense to optimize for performance, scalability and safety with languages like Java, C#, Go, etc.?

@jessesingal It's the unless

I genuinely don't know what to make of people who claim that current LLMs aren't "intelligent" unless the point is a very trite one about consciousness or the absence of wetware. It's this weird reflexive incuriosity about one of humanity's strangest and most fraught inventions.

How it feels to be working in AI at the current moment...

@sabinedoering EXACTLY. Can we put this on a t-shirt or something

This debate isn’t about whether AI is “smart” in some human, philosophical sense. It’s about functional capability. If a system can perform complex cognitive work reliably and cheaply, questions about consciousness or “real understanding” miss the (economic) point.

Ok, big time power up: #Codex in the terminal. I know, "duh" - right? Well, I've been using the Codex App, the VS Code plugin, and had just become accustomed to certain "quirks" (slowness, lack of visibilty into what its thinking/doing. Yesterday, I fired up the terminal, turned on the experimental "Parallel Agents" feature and... BAM. I feel like I just unlocked a whole new level of cognition.

@TheZvi parallel agents (terminal only) are - literally - a whole nother level.

GPT-5.3-Codex reaction thread. How goes the codexing?

@soundslikecanoe definitely. For me, its the non-deterministic nature of it all. It feels like one second things are working beautifully but you could wake up one morning and its all going haywire.

I'm not usually an anxious guy But lately my heart rate / dopamine levels have felt through the roof It's hard to say but it feels connected to all the constant AI stuff I'm working on. Constant context switching and information overload. Anybody else?

@mattshumer_ Dang. This is 🔥

@Rightanglenews This is actually fire

BREAKING - A man is going viral after translating the song Bad Bunny performed during the halftime show, and however bad you think it sounds in Spanish, just wait until you hear the lyrics in English.