# Don't get locked in — Kogen [Notes](https://kogen.dev/notes/) # Don't get locked in Almir Sarajčić · 26 October 2026 For more than 20 days now, my Codex usage limits have been resetting unpredictably. I don’t know if I’m getting a reset today or when it will land, and maybe once did one land when I actually needed it. That ruins my workflows and planning, and I don’t want to invest so much into one harness that I can’t take my skills, instructions and setup to another one when I’m not satisfied with it. [!\[Almir Sarajčić\](https://pbs.twimg.com/profile_images/1861769506642604032/AYcB3yTk_normal.jpg)](https://twitter.com/ultrathinktrash/status/2114594000896176329) [Almir Sarajčić](https://twitter.com/ultrathinktrash/status/2114594000896176329) [@ultrathinktrash](https://twitter.com/ultrathinktrash/status/2114594000896176329) ·[Follow](https://twitter.com/intent/follow?screen_name=ultrathinktrash) [Replying to @thsottiaux](https://twitter.com/thsottiaux/status/2114588174249071102) Hopefully in the morning. I can't spend all this constructively in an hour. [!\[Image\](https://pbs.twimg.com/media/HViL8ZfXwAAWZzG?format=png&name=small)](https://twitter.com/ultrathinktrash/status/2114594000896176329) [5:44 AM · Oct 26, 2026](https://twitter.com/ultrathinktrash/status/2114594000896176329) [2](https://twitter.com/intent/like?tweet_id=2114594000896176329) [Reply](https://twitter.com/intent/tweet?in_reply_to=2114594000896176329) [Read more on X](https://twitter.com/ultrathinktrash/status/2114594000896176329) This morning, 26 October, I was planning how to spend what I thought was my last hour before the next reset when I found out it had already reset. I only noticed because I kept checking my usage, and I still hadn’t seen confirmation that it had landed. Around the same time I got “Selected model is at capacity. Please try a different model.” [!\[Almir Sarajčić\](https://pbs.twimg.com/profile_images/1861769506642604032/AYcB3yTk_normal.jpg)](https://twitter.com/ultrathinktrash/status/2114639196694700059) [Almir Sarajčić](https://twitter.com/ultrathinktrash/status/2114639196694700059) [@ultrathinktrash](https://twitter.com/ultrathinktrash/status/2114639196694700059) ·[Follow](https://twitter.com/intent/follow?screen_name=ultrathinktrash) Guys, the reset already landed 2 hours ago. You're just wasting your new fresh weekly allowance. [!\[Image\](https://pbs.twimg.com/media/HVi0zIwWIAA0g02?format=png&name=small)](https://twitter.com/ultrathinktrash/status/2114639196694700059) [8:44 AM · Oct 26, 2026](https://twitter.com/ultrathinktrash/status/2114639196694700059) [0](https://twitter.com/intent/like?tweet_id=2114639196694700059) [Reply](https://twitter.com/intent/tweet?in_reply_to=2114639196694700059) [Read more on X](https://twitter.com/ultrathinktrash/status/2114639196694700059) I got this again while doing research on harnesses and writing this Note. My hypothesis is that people are ramping up consumption because they expect a reset, unaware it has already happened. They’re spending their new quota thinking it’s what’s left of the old one. With a lot of people doing that at once, the model could end up at capacity. I keep checking my usage because I don’t know whether a reset has landed. Meanwhile, expecting another one rewards burning through tokens instead of using them wisely. I don’t want my work to depend on those resets continuing. [!\[Almir Sarajčić\](https://pbs.twimg.com/profile_images/1861769506642604032/AYcB3yTk_normal.jpg)](https://twitter.com/ultrathinktrash/status/2100991421104685491) [Almir Sarajčić](https://twitter.com/ultrathinktrash/status/2100991421104685491) [@ultrathinktrash](https://twitter.com/ultrathinktrash/status/2100991421104685491) ·[Follow](https://twitter.com/intent/follow?screen_name=ultrathinktrash) I was just looking to see if someone wrote about the link between intermittent reinforcement and random usage limit resets - how we got addicted to Codex. This guy explains it well. [!\[Aakash Gupta\](https://pbs.twimg.com/profile_images/2021355466216062976/8MDXp7vR_normal.jpg)](https://twitter.com/aakashgupta/status/2091010976317661590) Aakash Gupta @aakashgupta OpenAI turned a rate limiter into the best growth loop in AI. Codex went from 3M users in April to 20M this week, adding 5M in the last 9 days alone, and the engine is a reset button one guy presses from his desk. The loop is simple. Usage limits create scarcity, and scarcity [4:52 PM · Sep 18, 2026](https://twitter.com/ultrathinktrash/status/2100991421104685491) [1](https://twitter.com/intent/like?tweet_id=2100991421104685491) [Reply](https://twitter.com/intent/tweet?in_reply_to=2100991421104685491) [Read more on X](https://twitter.com/ultrathinktrash/status/2100991421104685491) Until summer, I was using Claude exclusively. Then I tried Codex, maybe because of Claude’s usage changes, I can’t remember, and I liked it so much that I was almost exclusively using it. I downgraded my Claude subscription. When Opus 5.5 came out, I started using Claude more again, and now I’m balancing between the two. [!\[Almir Sarajčić\](https://pbs.twimg.com/profile_images/1861769506642604032/AYcB3yTk_normal.jpg)](https://twitter.com/ultrathinktrash/status/2101280681842053472) [Almir Sarajčić](https://twitter.com/ultrathinktrash/status/2101280681842053472) [@ultrathinktrash](https://twitter.com/ultrathinktrash/status/2101280681842053472) ·[Follow](https://twitter.com/intent/follow?screen_name=ultrathinktrash) Got that natural reset. First thing I'm working on: fixing my workflows to not rely on ChatGPT anymore. I'll finally try cheaper LLMs from alternative providers that I'd falsely believed were insufficient for daily tasks. We're not obligated to stay with Anthropic and OpenAI. [!\[Image\](https://pbs.twimg.com/media/HSk-BH0XAAAVcNn?format=jpg&name=small)](https://twitter.com/ultrathinktrash/status/2101280681842053472) [12:02 PM · Sep 19, 2026](https://twitter.com/ultrathinktrash/status/2101280681842053472) [1](https://twitter.com/intent/like?tweet_id=2101280681842053472) [Reply](https://twitter.com/intent/tweet?in_reply_to=2101280681842053472) [Read 1 reply](https://twitter.com/ultrathinktrash/status/2101280681842053472) What I’d rather not redo every time is what I build around a harness. If a skill depends on a tool only that harness offers, or my instructions assume its commands, those dependencies become work I have to deal with when I move. The more I put into customizations like that, the more work I’m leaving for the next switch, even if another model or subscription would suit me better. Claude subscriptions show how access and billing can change too. OpenCode removed its Claude Pro/Max subscription plugins in 1.3.0 because Anthropic prohibits them; Claude remains available there through API billing ([OpenCode providers](https://opencode.ai/docs/providers)). You can sign into an unmodified Claude Code yourself, while third-party products can’t route requests through subscription credentials on behalf of users ([authentication rules](https://code.claude.com/docs/en/legal-and-compliance)). Anthropic also announced that Agent SDK and `claude -p` usage would move off subscription limits, then paused that on 15 June, so subscription use continued. On 7 October it added monthly API credits for eligible Max and Team plans while keeping subscription use ([SDK update](https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan), [API credits](https://support.claude.com/en/articles/17154008-monthly-api-credits-for-max-and-team-plans)). Our September subscription comparison covered different restrictions: Z.ai’s GLM Coding Plan is limited to supported tools, with a separate written agreement needed for direct use from your own applications ([terms](https://docs.z.ai/legal-agreement/subscription-terms)); Alibaba’s Coding Plan prohibits automated scripts and non-interactive use ([Coding Plan](https://www.alibabacloud.com/help/en/model-studio/coding-plan)); OpenCode Go supports OpenCode and other coding agents ([Go](https://opencode.ai/v2/docs/console/go/)). If another model or subscription fits me better, I want to move the skills and instructions for my work, then adapt the parts that depend on the old harness’s tools or commands, without having to rebuild the rest of my setup. [← All notes](https://kogen.dev/notes/) --- Canonical page: https://kogen.dev/notes/2026-10-26/dont-get-locked-in