One part that makes me wary of these tools is security.
If I use a remote MCP or CLI that relies on network calls, and I give it in the hands of my coding assistant, wouldn't be too easy to inject prompts and exfiltrate data from my machine?
At least MCP don't have direct access to my machine, but CLIs do.
There is another differentiator between CLIs and MCP.
The CLI are executed by the coding assistants in the project directory, which means that they can get implicit information from there (e.g. git branch and commit)
With an MCP you would need a prepare step to gather that, making things slower.
I wonder if flood and drain would work with orchids.
I do that manually with my plants twice a week, they have flowers almost all year, but it's a chore to bring them out, flood them, make them drain and bring them back home.
Also my wife always yells at me because I always wet the floor in the process.
In my experience performance of LLMs can be surprisingly good on things that are not mainstream, like database engineering, and surprisingly bad at mainstream categories approached in an unconventional way.
That said, I'm amazed that you have 50 years of experience and still able to have the mental flexibility to adapt to new development paradigms.
As you imply, this stuff isn't simple to pick up, and is completely different on how we have done our job without AI.
> Fine-grained permissions and policies. Not just what tools an agent can access, but what it can do with them. Read email but not send. Access one repo but not another. Spend up to a threshold but no more.
If nailed this is going to be interesting.
All the other solutions I've been sumbling around are either very hard to customize or too limited.
Docker sandboxing is kinda nice, but not enough to trust an LLM even with my messaging accounts.
I wonder how much of the Rollup bundling magic has been ported to Rolldown.
One thing that always made this kind of switch to Rust has always been that Rollup has become so sophisticated that's hard to replace with something new.
I think that their main problem is that they don't have enough resources to serve too many users, so they resort to this kind of limitations to keep Claude usage under control.
Otherwise I wouldn't be able to explain a commercial move that limits their offer so strongly in comparison to competitors.