I actually ended all govt contracts I was working on after this and ended my clearance forever. This was my "I can't keep working for them" moment. One of the purest examples of the US operating on ISIS levels of morality that exists.
Killing someone for absurd reasons then lying about it because while knowing it's an absurdity/mistake risking it happening again seems almost worse than believing/buying into an extreme/insane worldview and killing someone for it.
Already happens in my experience. I'll try to figure out why it keeps doing this one thing and it turns out it's from some poorly advised info it put in a markdown file 20 commits ago that CLAUDE.md or AGENTS.md tell it to treat as gospel.
So it beats Fable 5.1, by quite a bit, on every metric? Interesting.
Might have to use my $20 Claude sub some more. I was moving away from it to a $100 OpenAI one to avoid the Claudese and poor token efficiency of Opus 5, given that I couldn't use Fable 5.1 with my tier, but this is worth trying out.
Why can't they let 20usd claude subscriptions access fable in CC, as openai allows you to use astra and max modes in codex - you just pay for it in more token use.
People often use a bunch of subagents, poor context management (though Codex's tight context limits and constant compaction tend to mitigate this), a bunch of projects at once, etc., as well as not using workflows that do heavy planning once up front and then consult it rather than thinking endlessly about what to do during the implementation part.
It's also the case that working on massive codebases is just a different beast. If they've been slopmining a monorepo for months with 200x, then their codebase is probably Lovecraftian at that point and requiring extensive effort to iterate on.
I had to go to another data centre just outside London about 10 years ago with a colleague who had never been inside one - we had some colocation kit hosted there and were going to spend most of a day swapping some cards and checking things through the KVM. He looked at me really weirdly when I offered him ear defenders on the way there. He wasn't ready for the sensory overload - the noise, the heat, the movement of air, the artificial light, the slightly weird smell, it was all a lot. 3 hours in, he wanted out. Couldn't blame him, I wasn't far off myself.
Adjacent, I work in a lab so we have dozens of ovens and freezers running constantly, so I'm intimately familiar with the constant hum, even though my office is a couple of doors away from the oven room.
We had a full site shutdown last week, which involved all power being shut off, which meant no running ovens.
The lack of sound felt so strange, like I was somewhere I shouldn't be, or in some sort of liminal space. It was unsettling in the silliest way.
Honestly just getting it integrated into mobile phone swipe keyboards would be a godsend. If I type "We need to get going " and then swipe the word "now", I really do not think "mower" should be the word it chooses. Present a set of swipe-based likely words and the preceding text message to Jev, or similar model, and pick its highest prob word.
Watermarking has referred to this "spy" use case for quite some time. Digital items purchased for download often have them, for example. Even before the rise of digital downloads, screeners for movies had them.
Yeah I use my 256gb Mac Mini for iOS dev, which has always had enough space for an iPad and iPhone simulator with plenty to spare, but this eats way into that.
You hobble the expressiveness of the LLM and reduce its capability.
Think of an agentic harness as like a kind of body for the LLM. It gives it primitive inputs (read_file, web_search or whatever) and primitive outputs (edit file, respond to user, etc). Give it a command line environment (in a locked down sandbox, with as few or as many tools as you prefer), and you've given it a toolbox. It can do a whole lot more, faster and more efficiently. It can compose tools together. It makes fewer transcription errors manually shifting data around. It can tame verbosity with good protections in the harness and access to grep, sed and awk.
It's really up to you how useful you want your agent to be.
If you don't have something running somewhere, you don't have an agent, you don't have a harness. You've got a token generator, an LLM from the 2024 era.
That's where you are completely wrong. The point of MCP is that you can have an agent and a harness without running raw CLI or Python commands. Very common for relatively lightweight loads that involve shuffling data around between APIs, often run in a tiny serverless task.
Sure, you have an LLM which can invoke functions. I will say that without storage and composition, you're asking for hallucination. LLMs are not deterministic and while they're good at regurgitating text - you can see how replies in an instruct model are structurally keyed off the question - they will rephrase, adjust, "correct" data they're schlepping from one call result to another call input. And one noisy MCP call and there goes your context.
You could build something with storage and composition out of MCP functions, but come on, have you seen how LLMs - particularly budget LLMs - try and invoke functions reliably? The amount of retries you have to hide, feedback you need to send back to the LLM about what it did wrong. Parameters get replaced with synonyms, arrays are passed for singular arguments and vice versa, structured inputs are flattened, etc.
So maybe you fine tune on interactions with your subset of MCPs, to improve reliability. But all you end up doing is reinventing a Unix-like command line, poorly.
Firecracker micro-VMs, gVisor, wasm sandboxes. There are ways to make this work that aren't heavyweight. Giving LLMs tools that they've seen how to use millions of times in training corpora just works better.
Exactly. A lot of people complaining about MCP are doing so because their only interactions with LLMs are via big batteries including code harnesses and don't understand what kind of (usually much more domain-specific) agentic systems are being built. For example, "CLI vs MCP" doesn't make any sense whatsoever if the agent doesn't have access to a CLI!
MCP suffers from its harebrained choice early on to load everything into context up front.
reply