From f92ef3cc1322cc425de969e590561902f7b1389c Mon Sep 17 00:00:00 2001 From: Jessica Greene Date: Tue, 19 Aug 2025 21:10:36 +0200 Subject: [PATCH] Create coding_assistants.md --- coding_assistants.md | 80 ++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 80 insertions(+) create mode 100644 coding_assistants.md diff --git a/coding_assistants.md b/coding_assistants.md new file mode 100644 index 0000000..6eea93f --- /dev/null +++ b/coding_assistants.md @@ -0,0 +1,80 @@ + +- developers with ranging experience, 20 years to fresh out of their studies +- consent for using AI tools +- transcription tools being turned on in meetings before folks join or without asking +- company version of chat gpt, +- system prompt as control layer +- using for thesis + - looking to see if a framework can be developed for approaching this new paradyme of programming +- often I didn't find it helped me +- I had a data analyst task and I decided to use github copilot (gpt4.1) + - made very ofen the same mistake which I had to +- I used it for a coding assignment + - it made correct code for the sample file but not for the actual file + - it was good at explaning Django + - luckily I knew enough to understand where it was going wrong +- will junior folks get more reliant on this and not understand and seniors maybe more efficient (because they know enough to make it work) + - hard to get internships + - also hard to get senior roles + - makes people stay longer in a role, this has a knock on for role rotation + - maybe AI creates new jobs, because AI will make many problems (ie cleaning up poor data pipelines, security) + - titles changing + - back to start up mode, need to be more agile than agile + - lots of hiring in pandemic, after this period alot of people were fired, this trickled down + - senior fired for three juniors + - many times where hiring for tech only requirement was to have a pulse + - ML engineers being hired but then doing other roles + - hiring pay for ML companies - but can end up +- no longer efficient to spend hours on hours to code something + - I ask it to explain the code in a learning.md file, any technical concepts I should know + - we need some strategies to keep junior engineers understanding the code deeper so they are engineers + - a talk from a bootcamp teacher, the AI tools help the students get initially faster results but they hit the point of deciding if it's for them or not later (time sink) +- I am not using any assistant or AI, I feel I will forget things if I rely on them, I see juniors using them and they don't understand, repeating the same errors, I tell them to not use it +- coding assistants + - cursor + - different models, shortcuts, comments, developing prompts + - tried gpt5 in launch week, prefer claude sonnet (more technnical) + - claude code in the terminal +- writing typed code - the assistant are very verbose + - never write code that requires 100% of your thought process to understand it or you can't debug + - pearl a write only language +- overwhelming reports of job loses + - can be concerning also for senior devs + - try to look at the positives, can iterate quick and learn higher level concepts very fast +- software engineers are the middle class, so getting rid of them is essentially removing the middleclass +- vibe code, you can now vibe to a MVP + - business model to fix these +- what problems are we fixing? + - are we causing problems to fix +- context coding over vibe coding + - ie claude.md or another file + - found it not following, it turned out there was a "bug" that told the model to only consider it if it considered it useful +- how we feel emotionally + - some frustration but looking at positive side helps (otherwise get too frustrated) + - I got it to write some code, but it was wrong I felt I had to fight with it to get it to understand what I wanted/where it was wrong + - often I approach it when I feel lazy +- how to isolate where the system runs + - agents can run commands + - security, lots of work to be done + - when will we see the first big breach, dependency issue + - asks to important many random libraries +- human to err, to really mess it up you need a computer +- coding review +- code authorship, who owns the code if the assistant wrote it or changed it + - comment that it was written by AI and include the prompt + - reward system + - line of code, commits, moving tickets over to done + - whatever is measured people will work towards that statistics + - hard for maintainers, code coming potentially from AIs and this creates over head to review PRs where no one replies + - lack of funding going into open source, particullary python +- waste of generations + - energy + - machine sympathy +- how is it for senior engineers + - right now I don't know what I need to do so I have to research as a junior + - know the boundaries of the tooling and how to get something from it + - same process for seniors, ie we all use search engines but seniors know better the key words to use +- better to train a junior than spend time getting the AI to do it +- will it get so good to replace + - gpt5 described as being a PHD + - if you are not a domain expert than you don't know the expertise you don't know