Used Gemini Code Assist in VS Code under Google AI Pro plan to review and greatly enhance documentation of two Powershell scripts repos

Overview

Over past few days, I did my first signficant software development related work using Gemini Code Assist extension in VS Code under Google AI Pro plan for one public and another private repository of Powershell scripts.
Overall, it has been a very positive experience. I was very satisfied with the review and significantly improved documentation of these repositories. Earlier the first repository which has a set of subfolders each with multiple PowerShell scripts had only a single README.md at the project root folder level. After this work that top-level README has been improved and new READMEs have been created for each sub-folder covering the PowerShell scripts there. The private repo did not have any documentation. Now it is quite well documented.

Past Free Tier ChatGPT and Gemini Usage

In the past, on free tier, I have extensively used ChatGPT as well as Gemini on web (chat) for assistance for the PowerShell scripts in the above two repos. I should also say that initial work on these PowerShell scripts was done by me when I was not using these AI assistants. But somewhere down the line, I started using them for improving existing scripts and developing new scripts.

In such free tier usage I faced key limitations of:
  1. Having to upload script files code to AI assistant. In case of Gemini, I could upload files whereas in case of ChatGPT I would copy-paste the code as uploading a file would lead to ChatGPT quickly closing that chat session and forcing me to open a new chat session where the old chat session's context was not available.
  2. I had to manually update the code in my files using the responses provided by AI assistant.
  3. I faced quite a few errors if I had long chat sessions or uploaded/copy-pasted multiple files code and made requests where AI had to look at all the files' code (perhaps due to context window size limitation dropping earlier parts of the chat session - sliding window effect).
  4. Due to above issue, I used to limit the requests to AI tool to small tasks and do the work step-by-step.
  5. I had to explicitly explain the file structure of the repo when that information was needed.
  6. Note that VS Code CoPilot free tier provided a great way to easily get AI to view and edit files in current project. But I would exhaust the monthly quota of the chat free tier in one or two days when I used it heavily for a project. So I did not use CoPilot chat much.
Despite above limitations in free-tier usage, I did find both ChatGPT and Gemini (free tier) quite useful in the above work.

Gemini Code Assist in VS Code under Google AI Pro plan

The Overview section has remarks on how Gemini Code Assist (GCA) in VS Code under Google AI Pro plan helped me. Given below are some additional details that I am trying to jot down quickly (I don't want to spend too much time on this).
  1. GCA was able to view and edit files in the current project in VS Code like CoPilot chat free tier. The big difference was that GCA under Pro plan allowed me to heavily use Gemini AI in it over 3 days or so without any quota exhaustion issue. Pro plan also has limits but they are bigger and further they are per day limits IIRC as against CoPilot's per month limits. I must mention here that I believe GCA chat (Gemini) under free tier too has per day limits and so may be superior in this regard to CoPilot chat free tier.
  2. GCA could 'ingest' multiple files quite easily when instructed explicitly. Even implicitly it would ingest a few files of the current folder. It would have a general idea of the project folder and file structure.
  3. GCA responses involving file changes whether .ps1 script files or documentation .md files would be presented within VS Code with Accept/Reject options for each file change (like free tier CoPilot chat). That was quite convenient and faster than having to copy-paste from AI tool web chat.
  4. By default my requests were handled by Gemini 3.1 Pro model which seems to be a significantly better model for coding than what I used to get with free tier Gemini web chat.
  5. At times when traffic was high on Gemini 3.1 Pro model, my request would be handled by Gemini 2.5 Pro. I would be informed of that.
  6. The code review quality - how well it understood the code and conveyed that to me, the (minor) bugs it pointed out, the performance and other improvments it suggested - was good. I think it was significantly better than what I would get with free tier Gemini chat or ChatGPT chat. The documentation it produced was also good quality. It still tripped up at times but that was much lesser than the trip-ups I have seen with free tier Gemini chat or ChatGPT chat.
  7. The responses would take some time, especially when it would report high traffic. 2 minute wait during high traffic was not uncommon. In contrast free tier Gemini chat or ChatGPT chat responses would be faster. But then I think the latter used a 'fast' model whose coding quality was not as good as Gemini 3.1 Pro or Gemini 2.5 Pro. GCA allowed me to change the model to Fast but even with Fast model, during high traffic time, responses could take a minute or two to come.
  8. I had to play a reviewer of suggested changes role as sometimes the suggested change would be not good enough and, at times, have some errors. But overall the quality of suggested changes was superior to my free tier AI tool past usage.
  9. I could have long chat sessions involving many files without having any concern of the AI tool getting overwhelmed. I think the 1 million tokens context window of Pro plan gave me that vital feature. In free tier AI tool usage, I surely would have hit a point where the chat was getting hard to continue due to many errors or AI tool not understanding my requests, forcing me to switch to a new chat. I did not have to do that at all with GCA for the work for these two repos.

Comments

Post a Comment