Harold Collins

[Harold Collins @masterskriptor]

@Krypius @unclebobmartin One workflow I tested used that methodology. It uses the best models to write the code for the entire slice and save it to a single file. Then i used free, and local, models to edit the files. It kinda worked. lol

[Harold Collins @masterskriptor]

@Krypius @unclebobmartin The goal is to have an agent write the context for your more expensive agent runs.

You can achieve much higher quality code by passing a coding agent only the skills it needs to write the code, the specs it has to build and let it spit the code out.

[Harold Collins @masterskriptor]

@Krypius @unclebobmartin You have all the extra context to show it how to write code, when it doesn’t need the skill to write a file, or create a GitHub issue, and the thirty other skills it won’t use.

You can almost fit Uncle Bobs entire book in that wasted context. lol.

[Harold Collins @masterskriptor]

@unclebobmartin @Krypius So I could do parallel work in separate layers, especially once the domain aggregate was complete.

This was dotnet c#.

The only issue I had left was with ef migrations continuously trying to run using the ui project before it had been updated to the new specs.

[Harold Collins @masterskriptor]

@unclebobmartin @Krypius I had issues like that with my orchestration agent telling my coding agents how to write the code. I ended up creating agents in between the orchestration and the coding agents. That ended up working because they focused the context to one layer at a time.

[Harold Collins @masterskriptor]

@unclebobmartin Interesting. Most seem to think that linear lines are more brittle. One fails the chain breaks.

So I split to two parallel agents only when one agent has to wait for the results of two or more and they dont depend on each other.

What I do is a simple form of graph engineering

[Harold Collins @masterskriptor]

@unclebobmartin It’s the Irish man’s stove!

[Harold Collins @masterskriptor]

https://t.co/e35TUp5E6p

[Harold Collins @masterskriptor]

@unclebobmartin Uncle bob, trust me when I say this, ai, as nice as it can be, is just as corrupt as we are. And lazy. As it turns out. Seriously, you should start with the Ten Commandments as the first part in your instructions file (Claude.md etc.) lol

[Harold Collins @masterskriptor]

Please comment on this if you use AI apps like ChatGPT, grok, Gemini, or google’s ask ai feature.

https://t.co/kaGHaouZEP

[Harold Collins @masterskriptor]

@WolfgangRichtEU Precisely. Which is why he is Elon Musk and no one has ever heard of you. Because your utopia would never have the chance to exist if it wasn’t for people like him. And if Elon did what you said your utopia wouldn’t last a month, and the world would be worse off than it is.

[Harold Collins @masterskriptor]

@Pirat_Nation No offense but this sounds like some propagandized bs. The only safety concern in this is if google is using the model to store your personal information. The whole carbon emissions point has lost its validity until I see Miami under water like they said it would be.

[Harold Collins @masterskriptor]

@_nakedeyes @unclebobmartin Did you say something about uncle bob being biased? After you lead off with a completely biased, totally unsubstantiated delusional rant. Sounds like politics to me. Just another blood sucking insect among many.

[Harold Collins @masterskriptor]

@AntiWokeMemes I never owned slaves. You just mad cuz u don’t own me.

[Harold Collins @masterskriptor]

@CassandraRules I think she meant to say we don’t understand her psychosis…

[Harold Collins @masterskriptor]

@LibOrNormal Yea. I don’t recommend you do that maam. Bad idea.

[Harold Collins @masterskriptor]

@TrumpsHurricane Bye