I started using AIs heavily 2-3 months ago. For the record, I’m writing down my current AI usages.
My current AI workflow
Here is my AI workflow. I mostly use free-tiers. I feel $20 a month for an AI service is a bit too high although I know that they barely earn profits with the pricing. The only exception is Gemini API because I needed to top up $10 just to start using it.
Brainstorming project ideas
I tend to ask same questions to ChatGPT, Grok, Gemini and DeepSeek because I want 2nd and 3rd opinions. They sometimes give me surprisingly good ideas. Other times, they assure me that certain ideas are novel and excellent, but there are already many similar github projects. It is important to doublecheck.
Coding assistant
I use Claude as my coding partner/assistant. I like its concise texts.
I’m not a fulltime dev and I usually code for fun and learning. To avoid it from giving me complete code from the beginning, I instruct like:
Problem-by-problem pedagogy
Don't give complete code first — guide toward the solution
Concise answers with links to official docs where relevant
When I need a sample code block, I ask Grok instead so I don’t distract Claude from the instructions.
I usually don’t hit its 5/6-hour usage limit but if a chat session gets long, I ask Claude to give me the context of the chat session, and start a new chat pasting it.
Also, I copy&paste only some functions and definitions that I’m working on. I think it’s a good practice not to give it all files in a project for two reasons:
- to minimize toke usage
- I need to grasp what I develop
After all, I develop for my learning and not for vibe-coding.
Quick research and Q&As
Grok’s web search capability is impressive. It rarely hallucinates and always gives me latest info.
Troubleshooting
I ask Claude first because it knows the context as my coding assistant. If it starts saying something off after a while, I ask ChatGPT and DeepSeek.
API
Currently, I’m only using Gemini because APIs are usually not available for free. gemini-2.5-flash-lite is extremely cheap and can be used for simple tasks. If things get complicated, I switch to gemini-3.1-flash-lite. With my two recent AI-related projects, I spent about 60 cents (only!) in total out of $10 I had initially paid. If I ever need a stronger one, I might use DeepSeek (strong but still cheap).
Pricing as of 6/1/2026:
| Model | Input | Output | Cache Read | Notes |
|---|---|---|---|---|
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 | $0.01 | Cheapest overall |
| Gemini 3.1 Flash-Lite | $0.25 | $1.50 | $0.025 | Balanced cost/performance |
| Gemini 3.5 Flash | $1.50 | $9.00 | $0.15 | Premium Flash model |
| DeepSeek V4 (Flash) | $0.14 | $0.28 | ~$0.0028 | Extremely cheap output |
Comments on AI products
Copilot
First, I started using Copilot because it was the one my company approves and there’s no limitation for the free tier. But after several weeks, I noticed that it was rather bad among available AI services. I’ve stopped using Copilot since then.
But today, I gave it a second chance if it could be a coding assistant. But again, I gave up within a few hours. It tried to force me to write down Pydantic-like class structure first. If I push back, it pushes back until I say “I decide, you don’t.” Well, coding with Copilot is not fun, which is the whole point for me to code.
Gemini
Last month, Google changed its usage limitations to compute-based, and everybody seems to be complaining about it. As I am a relatively light user and as Gemini is not my 1st tier AI, I haven’t been affected much.
With that said, I don’t think it was a smart move for Google. Why this move when rivals are trying to expand user bases with strategic pricing? I’m very curious how things will turn out.
I use all Web UI (and app), Google AI Studio and APIs. Web UI/app for casual conversations and AI Studio for tech Q&As.
Grok
I had an impression that Grok was always down. But as it turned out, it is quite useful (when it’s alive) on searching latest information. I think it’s better than Google or Perplexity.
ChatGPT
I have an impression that GPT-5.5 is the smartest among models available for free tiers. I often hit its usage limit, so I use it only when I need it.
DeepSeek
As for smartness, I think DeepSeek is the 2nd in my list. DeepSeek’s cost vs. performance is quite impressive. It’s just that I am a bit conservative using DeepSeek as the US administration might ban it in the future.
Being a Chinese model, it stops responding when I ask something about China, which I rarely do. It also starts talking in Chinese sometimes. I can just say “English, please.” No problems for technical use.