__erik

This is super cool! The brew install path in the readme doesn't seem to work

remote: Repository not found. fatal: repository 'https://github.com/frigadehq/homebrew-tap/' not found

I kind of buy that Chinese models should only be usable via western hosts, but completely shutting them off would be a massive self own.

Here's a lot of really nice features in this language (I really like the actor / nursery designs), but a GIL in 2026 feels like a massive weakness

This is cool, it reminds me of stumbleupon from back in the day

x402 as a standard will never have a fee.

x402 abstracts away gas! You can try here from a human use standpoint: https://www.x402.org/protected And you can see how to accept stablecoins or spend in the the examples https://github.com/coinbase/x402/tree/main/examples/typescri... https://github.com/coinbase/x402/blob/main/examples/typescri...

Blockchains are fast and and cheap now! Modern blockchains like Base, Solana, Sui (and many more) typically have block times <2 seconds, and a stablecoin transfer can cost as little as $0.0005 independent of the $ amount being transfered

its not that 402 isn't used, its that there is no standard response, which makes it hard for AI agents to pay for access. x402 provides a standard response format clients can programmatically use to create a payment, either using crypto assets like stablecoins, or in future, fiat methods.

>Given that this runs atop Payment Required, doesn’t this mean that each API request would involve an extra one or two data transfers?

Yes, assuming its the first time you've interacted with an API endpoint and don't have cached payment requirements.

>Is there a reason why you wouldn’t pay ahead of time? I just understand why you couldn’t buy a few dozen/hundred/thousand dollars worth of credits, and wait until it runs low.

This would require manually integrating each vendor, which is totally valid if your agent performs a lot of a single type of task, but we suspect over time there will be huge value in agents being able to dynamically select tools / apis they want to use to accomplish their tasks, and dynamically pay for what they use

>Does x402 prevent the double-spending problem?

This depends on the implementation on the underlying network, but basically the spending signs an authorization for transfer, and the merchant either settles that onchain themselves or delegates to what is called a facilitator that settles on their behalf. On EVM chains for the exact payment scheme this leverages EIP-3009 signatures

Gas is variable, but Coinbase currently subsidizes x402 transactions that go through our facilitator

x402 as a protocol has no fees, but the underlying network transactions are conducted on my have costs. Merchants can choose the underlying network transactions are conducted on that best fit their usecase. x402 also has the concept of a facilitator, which exists to abstract away the underlying payment networks. Many facilitators (including Coinbase's) subsidize the gas used for transactions.

(x402 author here) Thats actually exactly what the intention with x402 is. Merchants express to the buyer multiple ways that they accept payment and the buyer chooses the one they prefer. As of right now stablecoins on various crypto rails are the major form (largely because they're the easiest to start with from a development standpoint), but as Cloudflare indicated in their blogpost, they're proposing a scheme that uses deferred payment via credit card or ACH https://blog.cloudflare.com/x402/.

we will happy accept PRs the add support for any chains that can perform the payment flows in a safe, non-custodial way for the client, resource server and faciliator. Currently x402 works with any EVM chain, and we're working with several chains on integrating.

People never discuss company size along with this question.

If you're at a giant company, the answer is likely no, there's enough politicing and paperwork where the highest impact thing to be done by a manager is likely not coding.

If you're at a startup / smaller more nimble org in a big company, the answer is likely yes, if you've gotten to the point where you're a manager, in theory you're a very good engineer and you should spend your time coding, but on things that aren't on critical path. Bug backlog, experimental things with no hard deadlines, proof of concepts, all of these are valuable things. Leading from the front is also just generally good with smaller groups.

Also under discussed by people having these debates (typically managers), is not acknowledging how bad most managers are at coding, especially if their job hasn't required them to code in a while. I see all the time that managers look for any excuse not to code, because it would reveal to their team that they're at best an L4 level coder after being in management for 5-10 years.

This looks really nice! Curious if its running in production anywhere

This is super cool. Haven't seen such a pragmatic framework for composing local LLM action, especially in rust

Data policy, knowing what and how data is stored, would be good enough

This looks really cool and something I would use, but I'd need to see more on data retention and security. Just by nature of the product it requires sharing a lot of potentially very sensitive information.

Zora.co | Senior Infra Engineer | REMOTE or NYC | Full time | Visa possible

About Zora: * Zora is a decentralized market protocol for NFTs * Globally distributed team, remote first but with an NYC office * Focused on creators and artists * Great comp and equity * Amazing entrypoint for people who want to get in to crypto, no crypto / web3 experience required, Zora is a crypto native company!

About you: * Terraform / infra wizard * Enjoy startup vibes * Comfortable with AWS, Datadog, GCP * If you've got experience with data thats amazing * Interested in crypto / NFTs / Web3

Shoot me an email! erik @ zora.co

If anyone is looking to switch, I've far preferred using Insomnia to Postman https://insomnia.rest/

We're still ironing out a few things but unit tests for various functions then we have smol statistically representative datasets for each model. In CI we train a model on the small dataset (aim for <5 mins e2e) then have a a suite of model metrics we care about, tests confirm values are within acceptable bounds and the test values are pulled into the PR.

What has worked fairly well so far:

Models:

- Models are structured as python packages, each model inherits a base class

- base class has define how to train, and how to predict (as well as a few other more specific things)

- ML engineer can override model serialization methods, default is just pickle

Infra:

- Code is checked in to github, Docker container built each merge into master

- Use Sagemaker BYO container to train models, each job gets a job_id that represents the state produced by that job (code + data used)

Inference / deployment:

- Deploy model as http endpoints (SageMaker or internal) using job_id

- Have a service that centralizes all score requests, finds correct current endpoint for a model, emits score events to kinesis, track health of model endpoints

- A/B test either in scoring service or in product depending on requirements

- deploy prediction jobs using a job_id and a data source (usually sql) that can be configured to output data to S3 or our data warehouse

So far this has been pretty solid for us. The tradeoff has been theres a step between notebook and production for ML engineers which can slow them down, but it forces code review and increases the number of tests checked in.

Julia is great but it doesnt play in the domain of apps and servers

I'm really excited for this. I'm unaware of any mainstream language with first class support for differentiation, I think its going to be really interesting to see what people use it for out side of ML.

I've been using firefox on my personal computers and phone as my main browser for ~18 months. Honestly at this point I prefer it to chrome. On mobile the ability to use native extensions like ublock origin is amazing.

Also Canadian (Vancouver / Victoria), spend a lot of time in SF. This has also been my experience but I'm wondering if its just a big city thing in CAD.

Canadian services are typically good especially for urgent care things, I only have gone private on one occasion and that was for an MRI years ago when there were fewer of them available. I had a tear in my MCL and they wanted to see the degree (relatively low priority, it wasn't a full tear and I didnt need surgery). The wait for the MRI was ~6 weeks but so just paid out of pocket privately and had the results sent to my doctor. Took 2 days and he got what he needed.