@PaxiGalaxi@ayesha_fatiima You can train to solve puzzles or Leetcode tasks. Result will be the same - you will be really good at solving puzzles or Leetcode tasks.
And Leetcode has about as much ti do with real world software development as pizzle solving does.
@evvvvince@JorgeCastilloPr That’s not agents fault. That’s is a setup fault. You need to setup a learning system. If you rejected this nit, it should learn that and not report it again. Easely fixed.
Try my AI development factory. It the only app that you will need to develop with AI. It’s not another harness. Forget about IDEs as well.
It will drive your feature implementation from start to green PR, meanwhile giving you full insight into a run.
I have been working in it for 9 month already and it is taking its final shape (hi Destiny 2 fans :D ).
https://t.co/DrVvbYCxmp
I used to be a language lawyer. In the late 80s I _studied_ the ARM (Annotated Reference Manual for C++). I knew the language cold, inside and out. I could make it do backflips. I shipped a lot of code using it. A LOT.
I’ve attacked other languages with a similar intensity and depth. Until recently, that was simply the only way to become adept at building software systems.
I now consider the deep understanding of a language to be a supreme waste of time and effort. I don’t want to learn Rust, or Golang, or Typescript, or any of the thousands of other languages out there. I’ll leave that to my agents.
What I need now is to understand the high level gestalt of the system. I want to see the systems and subsystems. I don’t need to inspect the cogs. If I truly need to understand a detail at the cog level, I’ll interrogate my agents. I’m not going to pretend to be a language expert anymore.
@datachad That was one of the first things I worked on and it paid off. When I hit my Codex limits, I can just swap agent with Claude and click resume. I am so used to it now.
I think the next step in AI Driven Development is OBSERVABILITY, constraints and recoverability.
Terminals and chat windows are not able to give us that in the best way. Custom workflow harnesses is only a part of the full picture. A comprehensive UI that lets you know whats happening and lets you control the execution - is the other missing piece.
@thisguyknowsai Skills are not the way to go. Runtime should drive development, while agents implement what runtime asks them to.
https://t.co/rBFYCDh1WZ
GPT 6.1 Terra is like going 1 year back. I already forgot models doing this, but Terra constantly just stops.
This keep happening. I try to give it a chance, but it just infuriates me every time with this behavior.
Even GPT 6 Luna does not do this.
We brought Touch to the Mac!
From the all-new iPad mini to the 14-inch MacBook Pro and iMac... and now MacBook Pro with PROLED XDR Display and Touch, creators have new tools to do their best work yet.
@si_mon_ami@Solvairo@OpenAI Because the keep reducing the limits “to reflect cost reduction”. Whats the point if the model cost reduction? Its cheaper for them, but not for us.
Its not a news to most, but adding verification step to your agents, really makes a difference.
Example:
1. Do a code review
2. Verify the findings keeping in mind the task spec.
This alone removes so much false positives.
It surprises me how many people legitimately if opinion that current AI models are not smart enough. At least for coding, models are amazing at what they do. The best thing that could be next is ability to use them without worrying about limits.
@AltmanProp@2ar2betr99169@Solvairo@OpenAI Ok, I dont use prompts, I use my own harness https://t.co/nOBtMW8MMk . Prompts wont get you there for sure. You are right about that.