tcdent

I do something very similar. With some hacky string matching you can detect wether a window is available for reception and get past that.

I desire a more robust system which injects the events into the harness directly, but haven't found a way that works across systems.

https://github.com/tcdent/dotfiles/blob/main/bin/agent-messa...

https://frontier.com/shop/internet/fiber-internet/7-gig

7gig WAN is widely available residentially at a reasonable cost.

Gigabit / 2.5 Gbit connectivity is already obsolete. Any modern product must have 10gbe WAN with the hardware to back up NAT at that throughput.

I mean, to be honest, I'm the one who sets the precedent for most of the teams that I work with; that's one of my roles.

And personally, I don't necessarily follow a hard line with this. My Python API services run foregrounded in a terminal with uv run, but that doesn't mean that I'm using SQLite instead of Postgres. Any of the backend services I need that don't make sense to be foregrounded still run the same versions as prod containerized. And front end teams will tend to run a containerized API service since they're not modifying it as often.

Point being there is some conscious leeway here, but it's bounded and well-defined.

I would also add that the high-quality ones are quite expensive, and that I found incorporating an RJ45 switch with an SFP uplink was more economical for the cases that needed RJ45 (or gigabit) connectivity.

The concept of developing on a system which does not closely mirror your production environment died with virtualization and containerization over a decade ago. You will experience unforeseen consequences if your development environment does not emulate your production environment as closely as possible.

If you are implementing 10 GBE at distances less than 5-7 m, I highly recommend standardizing on DAC cabling. It removes the need for these kinds of conversions that create these kinds of heat signatures.

Fair point. I should have said "popularized in the modern software vernacular by Rust".

OpenClaw is an application, not a harness. Yes, it contains a harness, but it is a complete product.

When building an agentic workflow there are enough primitives that rewriting them from scratch every time makes zero sense.

What is a tool? How does the LLM understand the tool? Formatting a native function into a serializable input/output pattern makes sense to generalize and that does not need to exist repeated in everyones application code.

We use libraries to interact with the APIs themselves; nobody would say writing a spec-compliant API client was poor practice. Agentic harnesses are just one layer above: I need to call the API and I need to do it with certain expected conventions.

A builder pattern and decorators.

Yes, Python has decorators, but they're best used as "filters" that apply to functions or methods. Cache this, serialize the output of this function always, prepare this function to be used as a tool by an agentic harness. Not registration, not flow control. You may disagree but someone has to say it; FastAPI influenced the modern use of decorators far too much in the wrong direction.

Builder patterns are a Rust convention, because Rust has no named keyword arguments. A Python function already exposes a named contract. There is very little reason to ever to sequentially pass configuration parameters in chained method calls. If you need to add state that doesn't exist yet to a constructor or factory, that is not a builder pattern. That is registration. The one place where builder patterns should be tolerated is query builders. They iteratively build on a concept and having the additional "slot" for metadata (method name plus keyword arguments) is genuinely useful. Using methods which accept single parameter instead of keyword arguments is incorrect.

There are certainly edge cases where you want native USB and display, but after initial bring up, the device is on the network, and can be managed over the network.

So, sure, nerd out and add more hardware to your rack, but I need a physical keyboard and mouse attached to a machine in my rack like once per year.

Solve it and understand it; seems intuitive to me.

I don't understand how that contradicts my question.

Isn't the whole point of the field of mathematics in a theoretical sense the pursuit of formal solutions?

So, why would they be advocating for limitations on arriving at solutions?

Everyone's gonna frame this as "AI is dumb".

And, yes, the current tech is pretty dumb.

But this is a blatant misapplication of the technology in an obviously sensitive use case with an implementation that's so exploitable the people driving it have certainly never heard the term "jailbreak" once in their lives.

Reminds me of a consulting call that I had with a very large internet provider about their new agentic chat support system.

"We're going to start with the request routing layer and move that to AI agents, and then work though the individual services."

I thought it was a wild architectural decision that they would choose to roll every single action that the system handled through an experimental layer. My advice was to start with a safe, repeatable process to validate the effectiveness in the wild, and then expand in the same manner, bringing edges in as they had "solved" the individual implementations.

So, while this is almost the exact opposite of that, choosing a high-value target with real repercussions as their leaf implementation still baffles me. Step zero of any AI integration plan should be prioritization. Companies are routinely failing at this very simple, not-even-technical aspect.

Time will tell wether I find myself singing "Apple II Forever" in the shower.

Losing two weeks while you try out a candidates fit has way less cost than bringing the wrong person on formally and spending the next year debating wether they were the right choice or not with all of the associated "soft" overhead.

This is my preferred proposal to new contacts as well (I set it up as a contract so there's a little less red tape, but even people that pursue me for traditional employment afterword usually land on an extended contract).

Two things it solves: You get to evaluate me, my ability to deliver, and how I interact with your team and I either bring real value within two weeks, or I don't. I can tell you verbally I am an indispensable asset or I can show you; other people have ruined the verbal trust layer which is why this whole debacle exists in the first place btw.

And more importantly, but less communicated, I get to evaluate you. How your team works, the level of talent present, management's ability to keep direction, and wether I genuinely enjoy what you have for me to work on.

Fast is a Netflix product so the fact that you've even heard of it is in direct relation to the weight of the brand that launched it.

speedtest.net has been the first search result on Google for "speed test" for decades. Partly the boost of domain SEO and partly the boost of it being an effective exit node for searches for that term for that long.

(Nobody searches "ookla" and nobody is going to search your tier-3 .com)

Yeah these channels used to be respected in that way.

And then app developers discovered that hooks like "look what you missed" work on users and so now we all have to get them in the same category.

I'm sorry but if you write Python functions/methods in camel case I can't take you seriously.

They most definitely have a person to talk to. They're not the largest Google Cloud user by far, but they are large enough to have human account reps.

The interesting and yet-to-be-explained part is why google flagged the account?

Put all the timestamps you want in the post mortem about what you observed, but you haven't addressed the root cause.

The "this doesn't make sense" part of the story likely has a real explanation that nobody wants to reveal yet.

Article claims this primarily affects US tech companies. Then refuses to elaborate on who and how.

It really doesn't bother me. The persistent flags about how people think something was AI generated are far noisier.

Technically all of my replies are from an LLM, too; they all went through transformer-backed STT.

Appreciate the thoughtful reply.

Absolutely agree the deterministic performance-oriented mindset is still essential for large workloads. Are you expecting that this supplements a traditional vector/semantic store or that it superceeds it?

My focus has absolutely been on relatively small corpii, and which is supported by forcing a subset of data to be included by design. There are intentionally no conventions for things like "we talked about how AI is transforming computing at 1AM" and instead it attempts to focus on "user believes AI is transforming computing", so hopefully there's less of the context poisoning that happens with current memory.

Haven't deployed WVF at any scale yet; just a casual experiment among many others.

I appreciate the effort you put into mapping semantics so language constructs can be incorporated into this. You’re probably already seeing that the amount of terminology, how those terms interact with each other, and the way you need to model it have ballooned into a fairly complex system.

The fundamental breakthrough with LLMs is that they handle semantic mapping for you and can (albeit non-deterministically) interpret the meaning and relationships between concepts with a pretty high degree of accuracy, in context.

It just makes me wonder if you could dramatically simplify the schema and data modeling by incorporating more of these learnings.

I have a simple experiment along these lines that’s especially relevant given the advent of one-million-token context windows, although I don’t consider it a scientifically backed or production-ready concept, just an exploration: https://github.com/tcdent/wvf

Exactly this. It’s counterintuitive for most people, but the more complexity you add to the systems (the more organic they are), the more sustainably successful they become.

Everyone is looking for a simple solution, but simple solutions don't take into account human social dynamics.

Nah, taking the risk is even more fun when the thing you're modifying holds more value.

Chopping the fenders on a Porsche 911 to install a widebody kit does not have the same weight as rolling the seams on an Jeep Cherokee.

I don't think you're wrong, but if you really want to really re-think the approach, building an orchestration layer for Firecracker like every other company in the space is doing is probably not it.

This is a fantastic idea.

On a nonzero number of occasions I have priced the cost of running an inference server with a model that is actually usable and the annual cost is astronomical.