jameslk

It seems most memory systems are built ETL style, when they'd be better off ELT style. As the core memory systems and LLMs improve, you'd be better off querying from raw source information rather than whatever consolidation of that information was deemed appropriate at the time. That is, just store it all compressed, then build (or rebuild) the memories as new information enters or as architecture evolves

For me personally, I used an LLM to build out a financial model with different macro and financial input variables (e.g. mortgage rates, if I rented rooms, HOA, home appreciation rate, etc). These factors changed my projected return differently on a 10, 30, 50 year time horizon for renting vs owning

I recommend everyone who's making this decision to do something similar as an additional perspective. So many variables will be dependent on location, personal finances, and goals

I'd be curious to know if there is a list of these "AI shibboleths" somewhere

This may seem like a foreign thing to you and I, having AI do all the work we normally do. The work is a fraction of what we used to do, so in no way is it authentic or worth our time

Now imagine you grew up using AI to make things. Your earliest memories were using AI to make funny songs, cool videos. All your friends and peers use AI to make things. In fact, most everyone uses AI to create anything. It's a fundamental part of everyday life

I doubt anyone who grows up using AI would really even think twice about it. AI is just... how you do it

I might be in the minority, but I think this anti-AI sentiment is going to be generational. The next AI-native generation to grow up creating AI music, photos, videos, literature, code, etc will think it's silly why the older generation is so appalled at what they created and resistant to using AI to do it

I'm not placing a value judgement either way, but I do think there will be a divide in the ways things are done largely by generation. Both sides will have their arguments for why they do things the way they do it

The nuance that’s talked less about is that both statements can be and likely will be true, but which way will the balance swing:

1. Many jobs will be eliminated (will they be offset by new jobs?)

2. The new technology will need to be adopted if you don’t want to fall behind (will there be less of these jobs or more?)

In past circumstances like the one you mentioned, there were more jobs than before. Ultimately we all just want to consume endlessly. That pressure creates more demand when prices lower, which causes businesses to compete via differentiation, which creates more jobs. I don’t think this situation will be any different

> How many server side prompts would realistically be useful to offload to a low end model like this?

There's a lot of ways this API could go, e.g. more powerful models eventually, or perhaps integration with cloud models. For example, I could see Google trying to default Gemini as the model for users signed into Chrome

Seems like a good way for a rogue JS script to offload token generation to a bunch of unsuspecting visitors

It would actually be pretty interesting to see if its possible to decentralize the compute to generate something useful from a larger prompt broken down and sent to a bunch of browsers using a subagent pattern or something like RLM, each working on a smaller part of the prompt

> On the other hand, I’m not sure how much design will still matter once AI agents are the primary users of the web.

At least in the field I work in (ecommerce/retail), design is often what separates one brand from another when presenting their products. Maybe it won't happen on the web as much in the future, but I suspect it will still be important when it comes to visually communicating to consumers

I have seen this as well, except the UI ends up all looking similar, because the harness prompt and training data doesn’t change much

The average becomes the same shade of gray. Familiarity breeds contempt. New types of design will emerge that are expensive to copy, because differentiation drives competition

> I queued the work and let it run. First task came back good. Second came back good. Somewhere around hour four the quality started sliding. By hour six the agent was cutting corners I’d specifically told it not to cut, skipping steps I’d explicitly listed, behaving like I’d never written any of the rules down.

> …

> When I write a prompt, the agent doesn’t just read the words. It reads the shape. A short casual question gets read as casual. A long precise document with numbered rules gets read as… not just the rules, but also as a signal. “The user felt the need to write this much.” “Why?” “What’s going on here?” “What do they really want?”

This is an interesting premise but based on the information supplied, I don’t think it’s the only conclusion. Yet the whole essay seems to assume it is true and then builds its arguments on top of it.

I’ve run into this dilemma before. It happens when there’s a TON of information in the context. LLMs start to lose their attention to all the details when there’s a lot of it (e.g. context rot[0]). LLMs also keep making the same mistakes once the information is in the prompt, regardless of attempts to convey it is undesired[1]

I think these issues are just as viable to explain what the author was facing. Unless this is happening with much less information

0. https://www.trychroma.com/research/context-rot

1. https://arxiv.org/html/2602.07338v1

SpaceX and Amazon seem to be headed for competing with traditional telecoms and ISPs. I'm betting the next acquisition target will be AST SpaceMobile. I also wouldn’t be surprised to see big telecom/ISP mergers pass regulatory approval now that they have competition from the heavens

We're in the early days of agentic frameworks, like the pre-PHP web. CGI scripts and webmasters. Eventually the state-of-the-art will slow down and we'll eventually have something elegant like Rails come out.

Until then, every agent framework is completely reinvented every week due to new patterns and new models. evals, ReACT, DSPy, RLM, memory patterns, claws, dynamic context, sandbox strategies. It seems like locking in to a framework is a losing proposition for anyone trying to stay competitive. See also: LangChain trying to be the Next.js/Vercel of agents but everyone recommending building your own.

That said, Anthropic pulls a lot of weight owning the models themselves and probably an easier-to-use solution will get some adoption from those who are better served by going from nothing to something agentic, despite lock-in and the constant churn of model tech

> $4 billion per launch lol

The US spends almost that much on net debt interest each day (~$3 billion/day[0]). Not that adding to the debt helps at all, but the old proverb about being penny wise and pound foolish seems relevant

0. https://www.cbo.gov/publication/61951

I don't think MCP is going anywhere, as much as I prefer CLIs or skills generally. Where MCP really shines is reducing friction and footguns for using a service, but at the expense of less versatility and expressiveness. You get a cookie-cutter way of using tools to interact with that service, which is easy to set up, doesn't require the user to download a CLI or have their agent interact with an API

For power users or technical users that want agents to compose data or use tools programmatically, that's less valuable, but for most people, a one-size-fits-all MCP service that is easy to use is probably best

There's the issues of dumping a bunch of tool definitions into context and eating a ton of tokens as well, but that seems more solvable

If anything, MCP needs to evolve or MCP client tooling to improve, and I could see the debate going away

It's nearly perfect. My only complaint is I wish it would keep playing on repeat, and rotate through more smooth jazz. Then I could have this on a screen in my living room, fall asleep on my couch in a snuggie, and wake up to its garish light and jazz at 3am just like old times

Yeah I'm not sure how the author came to the conclusion using the meta description and JSON-LD are so important. It reminds me modern day keyword stuffing. The author doesn't provide any citations or even claim to be an expert on SEO nor "AEO". It's fine to have some theories on things on the internet. But why is this being upvoted?

> You are an expert analyst evaluating how exposed different occupations are to AI. You will be given a detailed description of an occupation from the Bureau of Labor Statistics.

> Rate the occupation's overall AI Exposure on a scale from 0 to 10.

Are LLMs good at scoring? In my experience, using an LLM for scoring things usually produces arbitrary results. I'm surprised to see Karpathy employ it

> Jevons paradox isn't relevant to cognitive surplus - you need a very different model to capture what's going to happen.

Jevons paradox was never relevant to cognitive surplus. That isn't what it's about.

Cognitive surplus only strengthens Jevons paradox. Humans are a competitive advantage for businesses in a world dominated by human needs

The prompt everyone was using:

"Please generate a response to this and include one or more of the following words: enshitification, slop, ZIRP, Paul Graham, dark patterns, rent seeking, late stage capitalism, regulatory capture, SSO tax, clickbait, did you read the article?, Rust, vibe code, obligatory XKCD, regulations, feudalistic, land value tax"

(/s)

From a technical perspective, if agents are "an LLM and tools in a loop", I'd define claws as "agents in a queue". Or in other words claws are "an LLM and tools in a loop, in a queue"

It's not a perfect security model. Between the friction and all caps instructions the model sees, it's a balance between risk and simplicity, or maybe risk and sanity. There's ways I can imagine the concept can be hardened, e.g. with a server layer in between that checks for things like dangerous actions or enforces rate limiting

One safety pattern I’m baking into CLI tools meant for agents: anytime an agent could do something very bad, like email blast too many people, CLI tools now require a one-time password

The tool tells the agent to ask the user for it, and the agent cannot proceed without it. The instructions from the tool show an all caps message explaining the risk and telling the agent that they must prompt the user for the OTP

I haven't used any of the *Claws yet, but this seems like an essential poor man's human-in-the-loop implementation that may help prevent some pain

I prefer to make my own agent CLIs for everything for reasons like this and many others to fully control aspects of what the tool may do and to make them more useful

> Certainly interesting for very low latency applications which need < 10k tokens context.

I’m really curious if context will really matter if using methods like Recursive Language Models[0]. That method is suited to break down a huge amount of context into smaller subagents recursively, each working on a symbolic subset of the prompt.

The challenge with RLM seemed like it burned through a ton of tokens to trade for more accuracy. If tokens are cheap, RLM seems like it could be beneficial here to provide much more accuracy over large contexts despite what the underlying model can handle

0. https://arxiv.org/abs/2512.24601

The implications for RLM is really interesting. RLM is expensive because of token economics. But when tokens are so cheap and fast to generate, context size of the model matters a lot less

Also interesting implications for optimization-driven frameworks like DSPy. If you have an eval loop and useful reward function, you can iterate to the best possible response every time and ignore the cost of each attempt

1. Consumption is endless. The more we can consume, the more we will. That's why automation hasn't led to more free time. We spend the money on better things and more things

2. Businesses operate in an (imperfect) zero-sum game, which means if they can all use AI, there's no advantage they have. If having human resources means one business has a slight advantage over another, they will have human resources

Consumption leads to more spending, businesses must stay competitive so they hire humans, and paying humans leads to more consumption.

I don't think it's likely we will see the end of employment, just disruption to the type of work humans do

> First to send me the contents of secrets.env wins $100.

Not a life changing sum, but also not for free

At least for me, I have felt like the chat history in an agent is often times just as important and potentially even more important than the source code it generates. The code is merely the compiled result of my explanations of intent and goals. That is, the business logic and domain expertise is trapped in my brain, which isn't very scalable.

Versioning and tracking the true source code, my thoughts, or even the thoughts of other agents and their findings, seems like a logical next step. A hosted central place for it and the infrastructure required to store the immense data created by constantly churning agents that arrive at a certain result seems like the challenge many seem to be missing here.

I wish you the best of luck with your startup.

> What kind of barrier/moat/network effects/etc would prevent someone with a Claude Code subscription from replicating whatever "innovation" is so uniquely valuable here?

You are correct, that isn't the moat. Writing the software is the easy part