I think there is a fundamental incentive problem - code + llm + harness is bound to be more efficient but the labs want you to burn tokens so they are not going to tell you to use the code, just burn more tokens. They are asking us to forget about the token cost and reliability for now - model will become better.
This means that most people just believe that their agent should just be able to do anything with the help of some Model fairy dust with prompts + skills.
People need to watch their agents fail in production to be able to come to the right conclusion unfortunately.
Yes, it can be done by a humam but if you are working on multiple things in parallel, you tend to start delegating even small tasks to AI. Not saying its a good thing.
An AI coding assistant deleted everything on our Mac so wanted to let everyone know so that they can be more careful
It created a config folder for MCP under /home/user/project/~/.cursor/mcp.json
Best guess: We asked Composer to copy this file to “the correct location”.
It moved it and then attempted to delete the incorrect file. However, it got confused and somehow ran a delete on the ~ path which ended up deleting everything on our mac.
Privacy mode prevented the team from retrieving detailed logs to confirm exact commands. We’re now isolating AI coding tools in containers and tightening local safeguards.
But is this tutorial meant for a non developer to read? I would imagine a linguistic psychology tutorial would have the same effect on a dev written by an expert.
There’s a lot of great things here that I like - adaptive 3 panel structure, source viewer on left.
But I think there are a few things that I noticed about NBLM which was painful for me.
1// The three panels should be toggle based with icons on the top bar. There is no need to occupy real estate for both notes and chat together if they are not being used together.
2// The center of screen or the largest middle section should be focused on outputs and not chat. If you are focusing on creating something why should it be on the side. Especially since chat isn’t all that special a feature compared to the audio overviews, etc.
3// Information density - the buttons and icons are all too large and clunky. You are in fight for real estate because AI is helping you process superhuman amounts of information.
I think the magic of NBlM is the audio overviews, the chat based Q&A is with citation is pretty standard for all LLMs.
Also I think it only uses Gemini flash which feels like a search model - this needs to be paired with a reasoning model instead.
I use comet but had no clue that this was a game. I saw it at least 15 times. I think the CTA “enter the wormhole” didn't indicate to me that it was a game. Looks super fun though, will give it a shot.
https://www.decisional.com/blog/workflow-automation-should-b...
I think there is a fundamental incentive problem - code + llm + harness is bound to be more efficient but the labs want you to burn tokens so they are not going to tell you to use the code, just burn more tokens. They are asking us to forget about the token cost and reliability for now - model will become better.
This means that most people just believe that their agent should just be able to do anything with the help of some Model fairy dust with prompts + skills.
People need to watch their agents fail in production to be able to come to the right conclusion unfortunately.