Openclaw sucks? Am I the problem??
I setup a cron job to summarize my emails using himalaya and it uses up like $0.25 on Claude 4.6 sonnet just to send me a pretty low quality summary of the last 10 messages. Maybe it’s my prompting, maybe I just don’t see the immediate value in agents. It is summarizing the emails but it’s not doing a great job at sorting through what’s important and what’s not. Maybe I need to give it examples of what to look for. Or pay an insane amount for opus or new gpt. Ideally id want to be able to sort through my junk mail and get notified when it’s an email from a real human being. I also don’t want to pay an unreasonable amount to do this.
What are some use cases that are actually mind blowing?
I’m a beginner vibe coder and I built an “agent” for poker (my own poker web app) that just sends jsons through open router and prompts it to respond back in a json based on the game state and using that json I have the game register the bet and the amount (or fold etc). Is openclaw actually way more innovative than that? I feel like it’s not and I have no formal education or experience… Maybe it becomes more useful as models become significantly more sophisticated but imo the marginal improvement can only be plateauing at this point right?? Please prove me wrong I want my mind to be blown…
edit: I improved my prompt and made it more explicit and it did do better but it was also very expensive. The real test is when i have actual emails that are important within the mix of all the marketing spam i get.