Peter Steinberger
speaker
1,527 appearances
1 recordings
1 series
first heard Feb 2026
last heard 12 Feb
Peter Steinberger’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 1 in all, peaking in Feb 2026 with 1.
Appearances
I think the main beauty is that models are really good at calling Unix commands.
So if you just add another CLI, that's just another Unix command in the end.
And MCPs, that has to be
added in training.
That's not a very natural thing for the model.
It requires a very specific syntax.
And the biggest thing, it's not composable.
So imagine if I have a service that gives me better data and it gives me the temperature, the average temperature, rain, wind, and all the other stuff.
And I get like this huge blob back.
As a model, I always have to get the huge blob back.
I have to fill my context with that huge blob and then pick what I want.
There's no way for the model to naturally filter unless I think about it proactively and add a filtering way into my MCP.
But if I would build the same as a CLI and it would give me this huge blob, it could just add a JQ command and filter itself automatically.
and then only get me what I actually need, or maybe even compose it into a script to do some calculations with the temperature and only give me the exit output, and you have no context pollution.
Again, you can solve that with sub-agents and more charades, but it's just like workarounds for something that might not be the optimal way.
It definitely was good that...
We had MCPs because it pushed a lot of companies towards building APIs.
And now I can look at an MCP and just make it into a CLI.
But this inherent problem that MCPs by default clutter up your context, plus the fact that most MCPs are not made good in general, make it just not a very useful paradigm.
There's some exceptions, like...
Showing 1301–1320 of 1,527 · page 66 of 77
← Previous
Next →