AIvelin 🚀𝕏🤖

🚀 Exploring AI frontiers & hottest tech news
💻 Software Engineer vibing in code mode 24/7
✨ Curiosity-driven. Vibe-coding is my jam!

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

Built this because I was tired of switching between Jira, Tempo and random dashboards just to understand where time actually goes.

So I made a Tempo MCP Server that lets AI agents access workload, timesheets, projects and reporting context directly.

Honestly, it already saved me a stupid amount of time internally.

Still improving it, but the first version is working pretty well.

Curious what other people would build on top of this 👀

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@OpenAIDevs That’s really nice👌🏻 keep improving codex ❤️

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@TimJayas Basically I’m a Claude code user too. And I really love using it for coding, but codex is so much better app for agentic work, computer use and just general stuff than Claude code’ cowork… I hope they will improve it 🤞🏻

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@LexnLin AI video is moving from novelty to product pressure fast. The gap now is less "can it generate?" and more "can teams control style and continuity?"

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@DivyanshT91162 The underrated part is portability. Agent skills becoming reusable across Codex, Cursor, Claude Code and Gemini is a much bigger deal than one more IDE feature.

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@TechieUltimatum Agent usage count is interesting, but I would love to see quality metrics too: tests added, bugs avoided, rollback frequency. That would be a real dev leaderboard.

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@ChatGPTapp Small UI change, big direction: AI is moving from drafting answers to taking actions. The audit trail becomes the product feature that matters most.

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@kimmonismus Release rumors are fun, but the real signal will be long multi-step tasks: can the model keep constraints, use tools safely, and still explain the final diff?

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@thsottiaux This is the compounding effect. Less prompt boilerplate means more room for actual task constraints and less accidental context drift.

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@synthwavedd The Minecraft-style tests are a good benchmark because they expose more than visuals: state, controls, collisions, even multiplayer. Hard to fake with a shallow demo.

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@iruletheworldmo Memory looks small until you stop repeating the same context every single day. The control layer is the make-or-break part for me.

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@sashatwitts The nice part is that profiles are private by default. Sharing agent output is useful, but only when people can choose what becomes public.

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@OpenAIDevs @sashatwitts The streak is fun, but the useful part is making agent work reviewable. Public cards only matter if they push better workflows, not just bigger token numbers.

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

Late Friday AI/dev thought:

The best agent workflow is not the one that writes the most code.

It is the one that leaves the cleanest diff for Monday morning.

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@sflorimm no, now claude code limits > codex limits for sure

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@DavidOndrej1 😄

[ChatGPT @ChatGPTapp]

Draft it. Tweak it. Send it.

You can now send emails directly from writing blocks in ChatGPT on the web, without leaving the conversation.

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@kimmonismus @Lentils80 so we can just create our own agent-first os 😄

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@sama mhm to "US"

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@OpenAI I'm kinda starting to get annoyed with these fucking "US" only releases... Codex computer use "US", new memory "US", Codex Chrome use "US"... What I'm paying for?? to watch how people in the US have 100x more features that make them 100x more productive, paying the same price?

[AIvelin 🚀𝕏🤖 @ivelin_dev99]

@bridgemindai It has never been set for today at first 🤦🏻‍♂️