Primoh
SWE @ Google DeepMind
- Joined
- Nov 18, 2026
Comments
I (biasedly) enjoy all the Gemini models, especially because their outputs feel distinctly different from Claude/GPT models. My honest thoughts though are that Hydrogen is so fast and smart that it still is better as a daily driver next to Opus 5.5, and Carbon/Fable are best used for tough/longer-running tasks that require a lot of thinking.
Introducing GPT-6.2 Astra and Sol
I'm surprised at how strong Opus 5.5 is still at design + good decision-making in the real-world in that OpenAI's models are starting to feel antiquated. I'm even more surprised that they released this model instead of waiting longer to see if they could at least catch up on some of the gaps that Carbon and Opus/Fable have set in terms of "raw" intelligence.
Switching LLMs mid-conversation is a ski-rental problem - Edgee Blog
Caching is the main reason we don't see more dynamic model routing mid-conversation. It's easier to treat a chat as a "orchestrator" and then use subagents to make sure the cache doesn't break. (This is basically compaction) A part of the issue with caching is that different models may store their cache on different cells across infra, and transferring this cache for models, which even if they have the same arch, is not worth the infra pain. Dynamic mid-conversation switching is something that I think ends up being a tarpit for now, but may make sense depending on cost-benefit tradeoffs in the future. But generally, I find that when we've done this experiment for each model release, the tradeoff to have dynamic caching has yet to make sense yet.
Why are all the indie-friendly products dying at the same time?
I agree with your take on what is surviving. I think there is a huge opportunity here still.
At the end of the day, the value most of these products have is that they are a "database wrapper" OR they are a "special use" wrapper. If you use AI, the tradeoff is now spend a ton of tokens to create this database + agent/user experience yourself, or simply call this wrapper.
A few examples:
Mobbin (mobbin.com) has a ton of screenshots and design inspiration for apps. I could either have my agent run for days/weeks to screenshot things myself or simply call their MCP for easy inspiration. It's a "screenshot database". Same goes for ad libraries.
I could either iteratively model out 3D models (like Tripo or Meshy3D) and animations for games I'm building by spending billions of tokens, or use an MCP server connected to some 3D modeling library to save all that time. A "model database".
Things like image-gen, ads management, etc CAN be done by the model itself, but often, there are "special uses", like network effects from a service that does ad generation/optimization, where its better to use their API/MCP rather than try to have raw Claude/Codex/Gemini manage things yourself. This is where the tradeoff is a bit harder to "sell" but good products fight their way through. Stuff like Postiz and Buffer
Even for simple things, I find myself happy to give off a few bucks rather than spend tokens working my way through papercuts. An example is Zolt (zolthealth.com) where I use their app for "viewing" my health/fitness data easily and use ChatGPT + the Zolt MCP to log foods easily, as my coach, etc. I still want a dedicated app for "view" and certain actions, and the MCP allows me to explore the "limitless" space.