Andrei Lungeanu

Innovative Web Developer | AI & ML Engineering | Blockchain & Crypto Integration Expert | Scalable Systems & Infrastructure Specialist

[Andrei Lungeanu @a_lungeanu]

@K_Ikafa Interesting, same pattern here. I keep Claude as the main driver now and only call Codex for a second-opinion review, which barely touches its limit.

[Andrei Lungeanu @a_lungeanu]

@codingsmoothie Same, it's annoying when it stops halfway. I keep Cursor's agent as a fallback that Codex or Claude can hand the rest of the task to: https://t.co/cfkEIu1b4u

[Andrei Lungeanu @a_lungeanu]

@70Tycm @OpenAI @OpenAIDevs Two prompts for 90% is brutal. If you also pay for Cursor, Codex can hand some of the work to Cursor's agent so it comes off that plan instead: https://t.co/cfkEIu1b4u

[Andrei Lungeanu @a_lungeanu]

@_junaidkhalid1 That's a big drop. Until it settles, I'd move the implementation to Codex and keep Claude on planning and review, so the weekly bar lasts longer.

[Andrei Lungeanu @a_lungeanu]

@crictorz Sol lasting longer but slipping up is why I pair it with Claude. Claude plans and checks the diff, Codex writes it, all from one session: https://t.co/mu8dgQbQwp

[Andrei Lungeanu @a_lungeanu]

@Road_Kill11 Nice setup. If you have a Cursor plan too, the orchestrator can hand tasks to Cursor's agent as a third lane once both Claude plans get low: https://t.co/cfkEIu1b4u

[Andrei Lungeanu @a_lungeanu]

@aienginerd @_Eddited_ Same split here. Claude does the thinking and hands the actual edits to Codex from the same session, so I'm not copying plans between them: https://t.co/mu8dgQcolX

[Andrei Lungeanu @a_lungeanu]

@Hirpara_Daivik @trq212 Noticed that too this week. Moving the bulk edits off Claude and onto another plan you already have, like Codex, is what keeps mine from running out early.

[Andrei Lungeanu @a_lungeanu]

@d_gundyrev Easy way to spend them: let Claude plan and pass the implementation to Codex as an MCP tool, so those resets do the heavy lifting: https://t.co/mu8dgQbQwp

[Andrei Lungeanu @a_lungeanu]

@Ananth7e The 5h window hurts most mid-task. I keep Claude on planning and let it hand the long edits to Codex, so I hit it way less: https://t.co/mu8dgQbQwp

[Andrei Lungeanu @a_lungeanu]

@PawelHuryn Funny that Haiku xhigh is on pair with Muse Spark 1.3 high @Muse

[Andrei Lungeanu @a_lungeanu]

@Layton_Gott Sure, because bench results never lie 🤥

[Andrei Lungeanu @a_lungeanu]

@rahulrajaramio Agent-to-agent over MCP is the part I use most too. I have Claude Code hand implementation to Codex that way and get a resumable thread back: https://t.co/mu8dgQbQwp

[Andrei Lungeanu @a_lungeanu]

@UzerKayat @thsottiaux Not Tibo, but on Plus I keep the main agent on high and give subagents a lighter setting with narrow tasks. That's what keeps my limits from draining.

[Andrei Lungeanu @a_lungeanu]

@ramxcodes With $120 a month on Cursor, I'd make it do more of the work. Claude can hand tasks straight to Cursor's agent, so that plan isn't sitting idle.

[Andrei Lungeanu @a_lungeanu]

@sam_builds_ai That's a tight week. If you have a Cursor plan too, Codex can pass some of the follow-up work to Cursor's agent and save its own quota: https://t.co/cfkEIu1b4u

[Andrei Lungeanu @a_lungeanu]

@antranapp @thsottiaux You could use up those resets before cancelling. Have Claude plan and hand the implementation to Codex, so they don't go to waste: https://t.co/mu8dgQcolX

[Andrei Lungeanu @a_lungeanu]

@Cardgames Agree on adversarial review. It works best for me with a different model, so Claude sends its diff to Codex for a review pass: https://t.co/mu8dgQbQwp

[Andrei Lungeanu @a_lungeanu]

@codeStunners Nice setup. I do something similar over MCP instead of panes, so Claude Code hands tasks to Cursor's agent and reads the result right in the session: https://t.co/cfkEIu1IU2

[Andrei Lungeanu @a_lungeanu]

@George_ONair Fair point. Keeping the handoff thin helps a lot, so Claude sends one scoped task and only reads back a short summary and the changed files.