Следующий этап Codex может выйти за пределы одного ноутбука

Read the Technical Guide 2026 has already started strong. In January alone, Moonshot AI open-sourced Kimi K2.5, a trillion-parameter model built for multimodal agent workflows. Alibaba shipped Qwen3-Coder-Next, an efficient coding model designed for agentic coding. OpenAI launched a macOS app for its Codex coding assistant. These are recent moves in trends that have been building for months. This article covers five key trends that will likely shape how teams build with AI this year. ## 1. Reasoning and RLVR Early language models like GPT-4 generated answers directly. You asked a question, and the model started producing text token by token. This works for simple tasks, but it often fails on harder problems where the first attempt is wrong, like advanced math or multi-step logic. [...] The result is a model that understands software engineering practices like project structure, dependencies, and debugging, and knows how to use its tools to complete tasks. When you give it a complex

| | | | | --- | | | baq 7 months ago | parent | prev | next (javascript:void(0)) I’ve been using frontier Claude and GPT models for a loooong time (all of 2025 ;)) and I can say anecdotally the post is 100% correct. GPT codex given good enough context and harness will just go. Claude is better at interactive develop-test-iterate because it’s much faster to get a useful response, but it isn’t as thorough and/or fills in its context gaps too eagerly, so needs more guidance. Both are great tools and complement each other. | | [...] | | | | --- | | | baq 7 months ago | parent | prev | next (javascript:void(0)) I’ve been using frontier Claude and GPT models for a loooong time (all of 2025 ;)) and I can say anecdotally the post is 100% correct. GPT codex given good enough context and harness will just go. Claude is better at interactive develop-test-iterate because it’s much faster to get a useful response, but it isn’t as thorough and/or fills in its context gaps too eag

This functions like garbage collection. Technical debt is like a high-interest loan: it’s almost always better to pay it down continuously in small increments than to let it compound and tackle it in painful bursts. Human taste is captured once, then enforced continuously on every line of code. This also lets us catch and resolve bad patterns on a daily basis, rather than letting them spread in the code base for days or weeks. ## What we’re still learning This strategy has so far worked well up through internal launch and adoption at OpenAI. Building a real product for real users helped anchor our investments in reality and guide us towards long-term maintainability. [...] This behavior depends heavily on the specific structure and tooling of this repository and should not be assumed to generalize without similar investment—at least, not yet. ## Entropy and garbage collection Full agent autonomy also introduces novel problems.Codex replicates patterns that already exist in the repo

Over the last year, we witnessed this transformation first-hand at OpenAI. For the first few months after Codex was released to the public, ChatGPT remained the default AI tool for work within OpenAI. Through August 2025, the average OpenAI worker spent less than 10% of their tokens on Codex. Now, every department, including non-technical departments such as Legal and Recruiting, uses Codex as their primary AI tool for work. This pattern reflects what we believe will be the future of work given the expanded capabilities and accessibility of agentic tools. [...] People use Codex for longer-horizon work. By May 2026, 80.6% of sampled individual users made at least one Codex request estimated to exceed 30 minutes of human work, 70.2% made one estimated to exceed one hour, and 25.6% made at least one Codex request estimated to exceed eight hours. Codex became the primary AI tool for every department at OpenAI. Engineering moved first, but Legal, Finance, and Recruiting crossed into Code

Abderrahim Y. 6mo Report this comment Codex is essentially just the o3-mini model. They've shifted it from the $20 plan to the $200 plan and are now marketing it more actively, the only 10X engineer is the Agentic flow and to have the 10X software developper you have to understand all dev needs to build the Agents roles for each layer at we build this Agentic AI framework and we will extend it we don't need more models to have good output in my opinion Like Reply 3 Reactions 4 Reactions Emmanuel Ezeokeke 6mo Report this comment I wish the United Nations could ban all AI companies from releasing any AI product for the next 2 months so that we can rest for a while 🤣🤣 Like Reply 2 Reactions 3 Reactions Szymon Stasik 6mo Report this comment [...] Towards AI 6mo Report this comment 2 things of Codex that are different from usual: 1. Codex is powered by Codex-1 (bad naming i know) a fine-tuned version of O3, this could means better use of code edit tool, a

faster than ever before. And in part that's because we can um now ship not just features faster, we can also refactor things faster. So after we launched the codeex app, which mind you is only 3 months old um and has had like ships every week. Um, we've had several refactors since and like the benefit of this is that we can have one or two people work on quite drastic refactors of how state management works or other things like that. Um, while still developing other features because we can and without having to do like a major code freeze for like two or 3 weeks because we can um quickly incorporate the changes back in and iterate on things. The other part that we're doing um though is to then take the the learnings that we have and continuously feed that flywheel to make sure that Codex [...] Um, so I've talked about a bit about some of these things already, but like between uh Chronicle, which can learn how you're using tools and how you work with things even beyond codecs uh and the