The entire industry, including its outputs that LLMs are trained on, hasn’t reconsidered what’s easy vs. hard or fast vs. slow. LLMs consistently recommend against code changes because they will take “a weekend”. No, Claude. You will do the work and it will take 20 minutes.
New layer of abstraction. Someone needs to understand how to build that old C library and reprogram that embedded controller, but it doesn't need to be me. Give someone infinite possibilities. Tell them they can have any software change they want. Everyone will have some good ideas. Some people will be head and shoulders above the rest. Being able to do anything (not that LLMs exactly deliver on that goal) doesn't mean you'll do something interesting. It doesn't mean you'll invent a product worth paying for, or impress many people, or make anyone particularly happy or inspired.
I think it might be appropriate if what we're talking about is a human connecting an LLM to something it can't natively access. Like something in the physical world - a piece of hardware, a different human, software the LLM can't access. As the human in the loop your role might be to mostly just shuffle information back and forth from this external resource. But depending on the situation there could still be quite a lot of interpretation and analysis performed by the human in the loop.
With today’s technology I’d still want a GPU driver developer guiding the LLM rather than some rando who is out of their element. But cutting down the exploration cycle time and giving the developer massive parallelism (have 10x agents exploring different hypotheses or features) is the real win. We don’t need to skip all the way to slop just to squeak out a little more effort savings.
Never had any issues using it for personal small scale development. I think the premise of OpenRouter as a magical way to fall back on error or select providers based on price/tok-per-sec/uptime isn’t delivered upon. But it’s an excellent way to simplify the experience of hopping between model providers with a single unified billing system. I mostly only use first party model hosts. Theoretically there are alternatives with higher throughput or lower prices, but it’s simplest to not worry about that optimization.
reply