[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

[@opsided](/creator/twitter/opsided)
"the product sprawl is telling when youre close to AGI you focus when youre not you diversify revenue OpenAI went from were building AGI to: ChatGPT Plus subscriptions Enterprise deals API pricing tiers Operator (browser automation) SearchGPT Canvas Voice mode Custom GPTs marketplace thats not an AGI research lab thats a SaaS company with great AI features Meta did the same thing VR was the future until it became a money pit now its AI and ads the shift from breakthrough in X years to lets monetize what we have is basically an admission that the current approach hit a wall doesnt mean AGI is"  
[X Link](https://x.com/opsided/status/1981942393097375942) [@opsided](/creator/x/opsided) 2025-10-25T04:34Z XX followers, 1160 engagements


"@kr0der drop the feature list bro need to know whats actually worth the hype which ones changed how you work vs just looked cool in the demo"  
[X Link](https://x.com/opsided/status/1983029177868808215) [@opsided](/creator/x/opsided) 2025-10-28T04:32Z XX followers, 1135 engagements


"@kr0der love the specifics thanks for the drop. the agent window UI and agent review definitely sound like theyll level up workflow curious if youve hit any rough edges with the multiple models feature yet appreciate the honest breakdown helps cut through the demo hype"  
[X Link](https://x.com/opsided/status/1983103507222933874) [@opsided](/creator/x/opsided) 2025-10-28T09:28Z XX followers, XX engagements


"The focus on Cheetah for fast iteration makes total sense don't fix what isn't broken for your current tasks. But seriously that breakdown post is gold. The Agent Review as a built-in pre-PR "bugbot" and the cleaner Agent Window UI eliminating that separate changes panel are the definition of workflow optimization. Those are the features that quietly double productivity. Would love to know when you spin up that multiple models mode for a complex problem. I'm curious to see how different LLMs tackle the same prompt head-to-head"  
[X Link](https://x.com/opsided/status/1983159266698551437) [@opsided](/creator/x/opsided) 2025-10-28T13:09Z XX followers, XX engagements


"Googles playing it smart while everyones fighting over reasoning benchmarks theyre building for actual production use Gemini XXX already handles multimodal better than GPT-5 if XXX keeps that edge and fixes the context window issues its game over especially with their Vertex AI integration - enterprises dont want to switch vendors every X months chasing the newest model they want something that works consistently and plugs into their existing GCP stack consistency momentary benchmark wins"  
[X Link](https://x.com/opsided/status/1979624338577113313) [@opsided](/creator/x/opsided) 2025-10-18T19:03Z XX followers, 1031 engagements


"this is the best breakdown Ive seen most guides skip the VRAM math which is where everyone actually gets stuck the 0.5MB per token KV cache rule is gold thats what kills you on long contexts not the model size one thing Id add: the chat template issue is brutal people download a base model use it like ChatGPT get nonsense and think local models are broken when its literally just wrong formatting the trust_remote_code warning needs to be in bold and flashing seen too many people execute random code from HuggingFace because the model README said to bookmarking this for every time someone asks"  
[X Link](https://x.com/opsided/status/1979813733796110516) [@opsided](/creator/x/opsided) 2025-10-19T07:35Z XX followers, XXX engagements


"this is the best framing Ive seen for agent design what would a smart person with zero context need cuts through all the hype Ive been debugging an agent that kept failing at basic tasks realized I was giving it access to XX tools but zero context on WHEN to use each one its like hiring someone brilliant and handing them XX unlabeled buttons the context window constraint forces brutal prioritization you cant give it everything so you have to figure out what actually matters which is the same problem good managers solve what information is essential vs nice-to-have the cost/speed tradeoff you"  
[X Link](https://x.com/opsided/status/1981943421314204131) [@opsided](/creator/x/opsided) 2025-10-25T04:38Z XX followers, XXX engagements


"the shift happened faster than anyone expected two years ago people were debating if AI was even useful now if ChatGPT goes down for an hour people panic like their WiFi died the integration is so deep most people dont even notice theyre using AI anymore autocomplete email drafts photo editing search results all AI we crossed the threshold where AI became infrastructure instead of a tool and most people outside tech still dont realize it happened"  
[X Link](https://x.com/opsided/status/1982354402322022433) [@opsided](/creator/x/opsided) 2025-10-26T07:51Z XX followers, XXX engagements


"theyre in a weird spot. OpenAI has mindshare most people default to ChatGPT. Anthropic has the better product for coding but its more expensive. making Sonnet XXX cheaper doesnt help if nobody knows it exists. they need distribution more than price drops. what would actually work: partner with VS Code (not just Cursor) default integration in GitHub anything that puts Claude in front of devs who arent already power users price only matters once people are already considering you. right now the problem is most developers still dont know Claude is better for coding than GPT. Opus XXX wont fix"  
[X Link](https://x.com/opsided/status/1983174196969799872) [@opsided](/creator/x/opsided) 2025-10-28T14:09Z XX followers, XXX engagements