Logani Bangun

Builder. Writer. Generalist. Harvard GSE grad.

[Logani Bangun @lpbangun]

@kushxg Yep when the general population still haven't figured out what the '/' sign does in a chat box, we might end up with a new ux/modality that replaces it. Hopefully something a little more elastic and self-improving?

[Logani Bangun @lpbangun]

@robjama Are you referring to a similar concept as Sequoia's Services as Software? Where users just "NotebookLM" something and it gets done? https://t.co/l5yAKgzvi4

[Logani Bangun @lpbangun]

@RhysSullivan Hermes for regular cron jobs. Hermes is overall under hyped for what it can do. But once you test, it feels a tad bit overselling it's capabilities.

[Logani Bangun @lpbangun]

@petergyang Luna high/max is decent for browser use, testing, and reports for me. Not as good as a subagents though.

If you're doing an e2e test with hot fixing then you might fare better with terra as subagents.

[Logani Bangun @lpbangun]

@blader This is such an underrated observation. This is even more apparent in underdeveloped to developing countries where sociologically people have been accustomed/educated to wait for orders.

There might be a widening technological gap between cultures that we're under addressing

[Logani Bangun @lpbangun]

@runsonai From my testing it's good when it's set as a task/worker agent. Should be paired with a model like 5.6 Sol for doing the planning and setting constraints.

[Logani Bangun @lpbangun]

@goodhunt Whatever this is I'm reserving it

[Logani Bangun @lpbangun]

@joshm Yep. Agreed. Was teaching my friend how to automate job applications and doing practice interviews with ChatGPT voice the other day. Showed her how to work with Codex projects and it feels like she just discovered fire. It's actually something else.

[Logani Bangun @lpbangun]

@CrushedBetween @OpenAI You'd be surprised.

[Logani Bangun @lpbangun]

@heywayfinder @OpenAI Idk. I actually quite like it.

[Logani Bangun @lpbangun]

@anah_sahh Mistral and Yann Le Cun's AMI Labs

[Logani Bangun @lpbangun]

@elshayib_ @GoogleAIStudio Damn that's harsh hahaha

[Logani Bangun @lpbangun]

@neerajjj6785 Chinese models (Kimi, GLM, DeepSeek) and open source harnesses like Pi.

[Logani Bangun @lpbangun]

@solopribuilds Ngl I use antigravity CLI for basic research and benchmarking for other apps out there. It's still decent for web search and very fast too. But, I wouldn't pay for it. Got my sub free from university.

[Logani Bangun @lpbangun]

@yacineMTB Cursor for their generous grok/cursor model limits. Probably Qwencloud for trying out the new Qwen 3.8 max models.

[Logani Bangun @lpbangun]

@VadimStrizheus SSH experience, in-app browser for testing, and especially ChatGPT voice are killer features. Claude's app doesn't even come close.

[Logani Bangun @lpbangun]

@awilkinson @sama The harness isn't as powerful overall. Codex models perform better in Pi or even Claude Code from my testing. It's a bummer because the codex app and features like ChatGPT voice is really good.

[Logani Bangun @lpbangun]

@PeteCapeCod @BHolmesDev Truly

[Logani Bangun @lpbangun]

@docreator @sattyyouneed Care to elaborate?

[Logani Bangun @lpbangun]

@ayaboch Their finance function and browser use is all I use them for now.