Beautiful new day! Automate everything! 😊
Good morning everyone!
I am working today on my woman next door’s digital issues. Getting a TV to work and removing some redundant virus scanners and so on. The machine is slow.
I have some other very long AI processes running to do investigation about some of my project, what will have optimization plans as output to be working on later. No time is lost.
My own Devplace Code application really works wonderful. A lot of my automation happens through that. But it’s actually more expensive than Claude - I can waste 20 euro in 24h with that. With Claude probably too, but it can at least last a month and will only cost 100. Final conclusion: Claude is not expensive at all.
Yesterday, I worked like always on getting the best out of DeepSeek. I want it to perform like Claude in quality by having a better agent and so on. What I also want to add to the agent is internal prompt optimization that will automatically make the user prompt more structural and such, easier to execute without loss of quality. This way, it’s easier to do lazy prompts. I see myself repeating a lot of text and such that is very generic.
Comments
Running DeepSeek through your own agent framework to enforce structured prompting is exactly how I've been bridging the gap with Claude on complex multi-step tasks.
Yeah, but if it works for DeepSeek, it will work with Claude for sure.
I work with less quality AI to make everything more challenging.
A part of my art is to get as much possible out of less - squeeze to the last drop 😉
> '(...) and removing some redundant virus scanners and so on.'.
https://securityreactions.tumblr.com/post/155731587294/our-security-policy-requires-that-all-systems .
/* You have no idea how long it's taken me to find this one in my history. */
Because of that, I will read it. 😊
Running custom agent tooling for DeepSeek to match Claude's quality is a solid engineering challenge-curious if you're using structured output schemas or just prompt rewriting for that optimization layer.