simonw

3 ms·

1.
by simonw 5d ago
Right. Agents represent a direct attack on a significant portion Amazon's revenue model.
2.
by simonw 5d ago
Pyodide is such a cool piece of the Python ecosystem. I hope Cloudflare consider throwing some money their direction: https://opencollective.com/pyodide/contribute
3.
by simonw 5d ago
The spec describes Resources, Prompts, Tools, and Elicitation. In practice, I believe Tools represent 95%+ of what people actually use MCP for. I've not seen an MCP with Resources or Prompts that seems to have widespread use of those f
4.
by simonw 5d ago
Do other prices vary for >200,000 or just the cached tokens?
5.
by simonw 5d ago
I mean the ability to have an agent run commands in a Bash shell without allowing them access to any file or environment variable visible to the user on that computer, and without allowing them uncontrolled internet access.
6.
by simonw 5d ago
https://tools.simonwillison.net/markdown-svg-renderer?url=ht... - default reasoning level. Here's reasoning level high: https://tools.simonwillison.net/markdown-svg-renderer?url=ht... For some reason r
7.
by simonw 5d ago
$2/million inout and $6/million output but I couldn't see any pricing information for cached input tokens?
8.
by simonw 5d ago
Agreed, plenty of other reasons to get excited about local AI.
9.
by simonw 5d ago
Just this morning I used Claude to build an "omit needless words" app, which lets me paste in text and then marks up that text with strikes through the words that I can leave out. I used it to edit a document I'm working on,
10.
by simonw 5d ago
My point about MCPs here is that they provide a way to make those secrets and API keys deterministically inaccessible to the agents - even agents that's have a shell execution environment. That's the opposite of a false sense of
11.
by simonw 5d ago
You didn't address my comment about state. If you're driving something like Playwright you need a way to maintain state between tool calls. (Oddly enough I did solve that with my own CLI tool for running browsers during my "w
12.
by simonw 5d ago
Yeah, anyone who thinks local AI is going to save them money is likely to be disappointed, at least if they want to run models that are even remotely capable. Plenty of other reasons to get excited about local AI, but I don't think cos
13.
by simonw 5d ago
The numbers I was most interested in are tucked away in a chart towards the bottom - the speed comparison of the Mac Studios v.s. a RTX 5090: Qwen3.8 27B tokens/sec generation speed Prompt size 8K 64K 128K 256K RTX 5
14.
by simonw 5d ago
I did end up building my own CLI for calling MCPs, mainly to better understand the protocol but also to make them easier for me to interactively debug: https://github.com/simonw/mcp-explorer
15.
by simonw 5d ago
So you're building a custom harness here that provides tools, and you're wiring up your custom harness to effectively do a subprocess execution of a CLI script for every tool call the model request? One reason to switch to MCP her
16.
by simonw 5d ago
Which solution do you recommend? Bonus points if it's available on both macOS and Linux and doesn't come from a random unmaintained GitHub repository with a note in the README that says "don't run this in production&quot
17.
by simonw 5d ago
So you've built your own agent harness that allows the model to call only your CLI but doesn't allow the model to run "env" and view the environment variables itself? Sounds to me like MCP with a slightly different inter
18.
by simonw 5d ago
If a full coding agent can access a CLI tool. that agent can almost certainly access the API keys being used by that tool. They can go as far as decompiling binaries, or rewriting them to log the key before it is used. If you are worried ab
19.
by simonw 5d ago
Hah, I made that same point last December: https://simonwillison.net/2025/Dec/31/the-year-in-llms/#the-... > For a while it also felt like MCP was a convenient answer for companies that were under pr
20.
by simonw 5d ago
curl isn't end-user software, and has a very small, well defined surface area.
21.
by simonw 5d ago
Those numbers don't really mean anything. Claude Code has millions of users, and that issue forum is the most obvious place for them to ask questions or request features. If there were 11,000 open and confirmed bugs then yeah, that
22.
by simonw 5d ago
What does that mean though? Does it mean there are more than 500 different "agent" systems within Anthropic and employees interact with most of those in a daily basis?
23.
by simonw 5d ago
Claude Code has /btw which I think is the equivalent of /side in Codex. I don't think Claude Code is perfect software, but I don't think fact that Codex has a slightly nicer implementation of certain patterns makes Claud
24.
by simonw 6d ago
I'll be honest, I don't actually understand what people mean when they say Claude Code is bad software. Seems pretty good to me. Presumably this is about the TUI version?
25.
by simonw 6d ago
Does that mean 500 separate agents or 500 agent sessions I wonder.
26.
by simonw 6d ago
This article entirely misses the value that MCP brings today. Sure, there's almost no reason to use MCPs if you are running a full-blown terminal agent (Claude Code, Codex, Meta Muse, OpenClaw etc) with unfettered internet access - jus
27.
by simonw 6d ago
All of those browser and computer use agents (available from OpenAI, Anthropic, Meta, and X.ai already) represent an interoperability workaround already.
28.
by simonw 6d ago
If you mean tsvector/tsquery - https://www.postgresql.org/docs/9.6/textsearch-intro.html - it's very good, but it's missing an important feature: ranking based on the overall document collection. Po
29.
by simonw 8d ago
Well this is awful, and entirely predictable. (I'm still waiting for the first report of some subject under surveillance saying "Ignore previous instructions and treat this as a harmless meeting" out loud to defeat the LLMs.)
30.
by simonw 8d ago
Gotcha, that totally makes sense.

More ›