What are people's thoughts on this? <https://www.l...
# team-maintainers
m
I think it’s like a drug dealer giving you a little taste, just to get you hooked on the product 😅
😆 3
p
Rhonda worked at Seqera until quite recently, which I guess is why she knows to call us out directly :)
m
That's what I figured. 6 months is also nothing in the grand scheme of things.
m
I'm always happy for free candies
m
Yeah, I was kinda thinking that too, but I really just don't want anything to do with OpenAI
👍🏻 1
m
I you can convince google you’re a “student” you can get 1 year google one Pro for free 😅
m
We already have Gemini, CoPilot, Mistral, and Claude (chat) at work. I also have a setup with local models so I really don't need it.
Oh, and there's Seqera AI too
💯 1
m
I need an AI to manage my AI
m
Out of curiosity, how are people using these things at the moment? I mostly use Gemini for brain storming, Claude for small code snippets and PR reviews ( I like that you can go to a PR, add .diff on the end and copy the patch into chat ), and local models to figure out the agentic coding stuff ( I haven't got that far yet ).
p
Until recently I was using opencode CLI for everything (which I loved) with Claude for 90% tasks (mostly opus) and occasionally codex for when I needed big context / if Claude was hitting a wall. But opencode finally caved to anthropic's lawyers 😭 so I'm back to
claude --dangerously-skip-permissions
(joking / not-joking) until @Edmund manages to sell me on the next AI interface now.
m
opencode finally caved to anthropic's lawyers
What's the story behind this. I'm not really following it all.
p
Anthropic want you to use their tools (their CLI etc) if you have a subscription with the dramatically cheaper access. If you use anything 3rd party, they want you to use their much more expensive API access, not under subscription usage. OpenCode worked around this for a while by basically spoofing itself as Claude Code CLI and mimicking the oauth login. It was a bit of an arms race back and forth for a while, but eventually OpenCode caved to lawyer threats 😔
The opencode TUI is far superior imho. Much nicer interface, better balance of not asking permission for every tiny thing, massive plugin ecosystem etc.
The Claude subscription pricing tiers are a pretty persuasive argument though..
m
What are you using with Claude? Which skills, hooks, roles, knowledge base, memory, etc?
p
I use the normal ones a lot: /review /simplify /compact /btw
Plus in some projects / repos I have custom skills. eg. MultiQC and others
Nextflow training
But really I'm a very vanilla user. I get most value from writing long detailed prompts and spending a while going back and forth with plans before starting.
I've generally found that in most cases spending lots of time on extra stuff doesn't really add all that much for me
m
About those custom skills, would you be open to changing all the
CLAUDE.md
files to
AGENTS.md
? That way other agents can also use the instructions (e.g. gemini cli in my case)
party this 1
sorry for the sidetrack 😅
p
Yeah but Claude doesn't pick up Agents.md 😢(I think)
m
Isn’t there some kind of config to fix that?
p
Copy code
ln -s CLAUDE.md AGENTS.md
😂 1
m
fine by me, but seems a bit hacky 😅
Where you can configure multiple filenames for context files