The problem with AI is that you can spend months aggressively learning every new model, tool, and workflow. Then one release drops and half of what you learned feels obsolete. So I’m starting to wonder if the people doing nothing, sitting on their hands might win in the long run. Everyone else will be burned out, exhausted, and weirdly fluent in tools that no longer matter.
Im not sure. I see massive gains but i never used any agents.
Ive started writing my own chat interface and orchestrator layer the last days.
Mainly because i wanted to use deepseeks latest models. And 4 days in its already pretty nice. Cost control. Highlight. SubAgents. Better Chat Control. File Uploads. Better performance than claudes and googles stuff. Agents spawning sub agents. Basically choosing provider, load models, import assign agents, with an unified param handler.
Im just trying not go the full agent route. I want to still navigate the codebase. I want small scoped minimal context, with small scoped tasks. You can let it read files, but it cant apply changes. You have to by hand.
Its weird but otherwise at some point you just lose understanding. You wont know what the codebase is doing. Where what is happening and which functions even exist. And you are probably not much slower and wont feel that overwhelmed. While knowing which files you have. Where they at. How the flow is
My favorite joke is that you almost need to be unemployed in order to try to keep up with AI right now! 😂
Unemployed and specializing in one AI focus 😂
The problem with AI is that you can spend months aggressively learning every new model, tool, and workflow. Then one release drops and half of what you learned feels obsolete. So I’m starting to wonder if the people doing nothing, sitting on their hands might win in the long run. Everyone else will be burned out, exhausted, and weirdly fluent in tools that no longer matter.
AI has another issue - It’s not always the pace, it’s the overhype and sensationalism nature of AI
Never knew I had ADHD until AI came around
Im not sure. I see massive gains but i never used any agents.
Ive started writing my own chat interface and orchestrator layer the last days.
Mainly because i wanted to use deepseeks latest models. And 4 days in its already pretty nice. Cost control. Highlight. SubAgents. Better Chat Control. File Uploads. Better performance than claudes and googles stuff. Agents spawning sub agents. Basically choosing provider, load models, import assign agents, with an unified param handler.
Im just trying not go the full agent route. I want to still navigate the codebase. I want small scoped minimal context, with small scoped tasks. You can let it read files, but it cant apply changes. You have to by hand.
Its weird but otherwise at some point you just lose understanding. You wont know what the codebase is doing. Where what is happening and which functions even exist. And you are probably not much slower and wont feel that overwhelmed. While knowing which files you have. Where they at. How the flow is