jacobheric

I asked them and they said the plan is to make the spec public. I'll leave it up to them to say when that will happen. I'll say the spec is pretty flexible/configurable. You can swap our payment providers, it does not have to be stripe. And you can configure who does kyc and you can do that yourself.

I'm no expert. But I think it's the same thing that makes it easier to use for humans. So, `stripe projects --help` and then `stripe projects catalog` is super easy for me to grok and use versus reading some api docs and curling some endpoints.

It's provisioning expressed as an API, but wrapped in a cli that agents can very easily use. Then anyone that implements the API is added to a marketplace of services. And the ability to pay for service is baked in because it's stripe. The end result is that as an engineer I can ask an agent to use the stripe cli to discover and provision, including paying for, a wide variety of services and it mostly just works.

Wow, the amount of reactionary doomsaying in this thread is offputting. I built the Stripe Projects agentic provisioning integration (what this post is about) at Inngest and when I saw it actually work in action I thought it was actually kind of magical. Basically, hey agent, use my stripe account to provision a pro plan at Inngest, build me an app and deploy it. The agent does it and logs me into the Inngest dashboard. Agentic provisioning seems inevitable given the way we work now, and deferring to Stripe for kyc seems a reasonable way to do it.

I'm building an AI DJ for Spotify called Listen to Luther. It's open source here: https://github.com/jacobheric/luther. Currently it responds to prompts with a list of songs that can be queued, played or added to playlists on Spotify. Next I'll give it proper memory and the ability to adjust and remix on the fly. I built it because I was frustrated by the mostly passive experience of Spotify's own AI DJ. There is no path to commercialization here as Spotify's TOS does not permit it (I asked them). So I host it for myself and a few friends and made it open source for anyone who's handy enough to run it themselves.

If you use AgentKit + Inngest then you can do all the things you normally do with the Inngest dev server like observe runs with AI metadata and rerun functions and rerun functions from steps with edited inputs. We do have a step-through debugger coming pretty soon here as well. Note you don't have to use AgentKit with Inngest though.

Thanks! Yeah we don't currently augment the result data only the arguments. But its on our TODO list as we need to to send back stuff like error info about pipelines runs. Prisma has type utilities for this, see the Result<> stuff at: https://www.prisma.io/docs/orm/prisma-client/client-extensio...

We sort of do this with WASM for just in time pipelines. We write pipeline rules in WASM...for things like detecting/masking fields...then we import and execute those wasm rules in a variety of language SDKs. As a sibling comment indicates, it's pretty difficult getting data in and out, but it's doable. See here for an example: https://github.com/streamdal/node-sdk/blob/main/src/internal.... We do this sort of thing in node, go & python and are adding other languages.

This is awesome, the UI looks beautiful.

Thank you! I wrote the UI! It's a pretty tricky UI stack as we update everything to the browser realtime in protobuf over grpc streaming (using grpc-web and protobuf-ts). There is a lot mapping we have to do to shape the data properly for React Flow so we do that server side in Deno before passing it along to the browser. We still have some optimization to do to keep the live tail view zippy, but it's a pretty solid foundation.

In the node-sdk, process pipeline is async , see: https://github.com/streamdal/node-sdk/blob/main/src/streamda... (I'm the author of that). I believe this is also the case for the python and go sdks as well. So you can call it asynchronously for passive observability.

However, we implemented the pipeline rules in wasm with the goal of keeping the overhead as minimal as possible. So you could also use it as more of a data security or governance tool and invoke the pipelines synchronously and mask or prevent sensitive data before passing it along.

Hey there! I wrote the node sdk for this, you can use the node-sdk to to execute wasm-based rules on your data and interact with the data in real time. With the wasm rules you can do things like detect and mask PII, etc. The node-sdk is here: https://github.com/streamdal/node-sdk. There are some minimal examples of these pipelines in the readme and examples/sandbox directories.

https://www.jacobheric.com

Been posting mostly photos, and very occasional thoughts, for the last 15 years.

While your experience is interesting and instructive in general, I think it's not completely relevant to the specific loans outlined here. These loans are issued directly by the SBA, from federal treasury funds, for the purpose of assisting businesses that might otherwise not survive the economic impacts of covid. For any small business in that situation, it seems a reasonable option.

IANAL, but the collateral terms as I read them are very specific and limited to business machines and such, not your house or your car.

Yes, I believe this is correct. This post is outlining the details of the economic disaster loans that are offered directly by the SBA and described in the link by the parent. The current legislation refers to these existing loans as 7(b). I believe the new stimulus bill, at least the senate version that I skimmed, deals largely with loans made by other financial institutions and referred to as 7(a). Those are covered here: https://www.sba.gov/partners/lenders/7a-loan-program/types-7.... Of course that information is now made out of date by the just passed legislation.

I my experience, at least for an individual, I think saying no is exactly the opposite of dangerous. I’ve been contracting exclusively for 10 years and said yes about 6 times and no dozens or hundreds of times. As long as you say no frankly, with a reason, and offer some names of people who can help, people are happy with you and will more often than not ask again in the future. Saying no, especially for bad fit, is one of the most important parts of success I’ve had in consulting.

I didn't even realize google had a podcast app. I just installed it and you're right, I can't subscribe to a podcast since I have disabled web activity history. That's a particularly egregious dark pattern.

Interesting. I never really had a firm grasp on the overlap in functionality of ublock origin and privacy badger. I threw privacy badger in the mix at some point as I like the EFF and wanted to give it a whirl. I haven't noticed any sites loading slowly, but I'll do a comparison on some of the sites I use and see if it's slowing things down.

I use firefox with ublock origin and privacy badger and I can't recall the last time I ran into a site that was broken because of it. But, I visit a fairly narrow section of the internet regularly so there might not be much overlap between what I browse and you browse.

I also use the multi-account container add on and the temporary container add on. This allows me to pin a few big sites to their own containers (google, amazon, etc) and open all other new tabs in temporary containers. This setup works great and appears to help keep firefox fast over time. I use duck duck go to search but firefox makes it trivial for me to re-run a search with google if I need to.

I also run an ad blocking vpn on google cloud using Algo. I use google cloud because the vpn can run on the permanently free tier and I only pay for network traffic (which is near zero), and I also enjoy the irony of it. I have wireguard clients setup on all of my devices to use the vpn either permanently (phone) or on demand (laptops). Having this vpn is nice as it makes it easy to block ads in apps on my kids mobile devices.

This vpn setup works ok but not quite as well as when I ran the same thing using Streisand and open vpn clients. I only say this because I have a homebrew whole-house audio setup with a bunch of google audio chromecasts and no matter how I tweak the wireguard client settings I cannot get that casting to work properly. With open vpn clients, those settings are a cinch.

I agree and think all the "shocking" and "bizarre" assessments are clickbaity. He did a similar thing against Karjakin. It's a smart strategic move given the situation and his advantage in the tiebreaks. Having watched the Karjakin tie-breaks, I can't wait. It's going to be so exciting. Having said that, if Caruana should defy expectation and prevail it will also be a great story and counterpoint to Carlsen's strategic calculation.

Abstract: http://www.nature.com/nature/journal/v543/n7646/full/nature2...

This looks to be the software used to analyze the data and suggest the new tree: http://www.lillo.org.ar/phylogeny/tnt/

Thanks for the link. Great documentary. I just watched that with my kids. We were absolutely mesmerized by the earnestness and openness of the family profiled.

I love that book. It's the reason I got back into writing as an adult. My wife gave it to me and I read it almost straight through and I couldn't believe that such interesting stuff was "Oprah" popular and marketable. I was heartened and started writing again. If you like it, Blood Meridian is even more harrowing.

I actually think that's not a bad way to write! Whatever gets you into the zone or flow. Of course, you have to have the discipline to go back and edit with clear eyes later.

There is a famous (mostly misattributed) quote to that effect: http://www.goodreads.com/quotes/84190-sometimes-i-write-drun...

Though Hemingway didn't say that, he was definitely a merciless editor of his own work. I dropped in on this (http://www.themorgan.org/exhibitions/ernest-hemingway) exhibit recently that had early drafts of some of his most famous stories and books and it was remarkable how many whole pages and chapters he chopped out just prior to publication.

Yes, I recently stumbled upon Black Wings Has My Angel by Eliot Chaze. Great, obscure and underrated crime/noir/pulp novel. It transcends the genre. It's somehow both more gritty and more literary than the more famous Hammett and Chandler and similar stuff.

Here's what I do. I usually start with a dream. When I wake up, if I've dreamt something interesting I write down a brief sketch or scene or even just an image in a google doc. Then I mull it over for a few weeks, adding new ideas and details to the doc. I love to use the time that I'd otherwise be using for insomnia. If I find myself awake in bed, I spend time tuning the idea or the scene until I fall back asleep. Then I write it down in the morning. Pretty soon I have a plot or a basic outline or just an idea for a beginning or an end. Then I continue to chew on it until a specific first sentence strikes me. Then I drop everything and start writing like a maniac. When I'm going on the writing, I try to let the characters live and let the story go where the characters take it, knowing that they'll eventually end up at the signposts I've set up in the initial sketch. I always use a trick that Hemingway used: stop writing for the day when you know what's going to happen next. That way you'll start right off again tomorrow and not get stuck.

I've written three books this way and I'm writing two more now. I've started a publishing company and published one so far. I'm working on ushering two more through that process (editing, editing, editing, typesetting, etc). That's a whole other story, and just as important.

EDIT: I have a day job (most of the time) as a developer, I do this in my spare time and between gigs.

I'll check it out. Thanks! Some of my favorite reading is genre fiction that trims the fat and floats above the rest of the genre. This seems to fit that bill.

I always thought King's writing could stand to be edited down (a lot). But I feel that about most genre fiction. I find it stifling as a reader especially when characters thoughts and emotions are over described. Usually a simple detail about their appearance, their manner, or similar is sufficient to clue me into what's going on inside. And the uncovering of those clues makes for a much more enjoyable reading experience.

If you consider a novel a work of art, then creating it is an act of artistic creation that has some intrinsic value. There is a benefit to expressing yourself carefully, to telling a story that says something about the world that you think needs to be said. Those outcomes are "significant" even if the novel has a small audience.

I start out in google docs because it's convenient and the change history is nice to have. I do this instead of plain text because I like adding some formatting as I write. If the book gets over 200 pages (or thereabouts) google docs starts to have problems and then I switch over to Pages on OSX. There, I save edits to google drive as I go. When it comes to (self) publishing time, I actually end up typesetting in word (for print) because it's the easiest, least horrible tool for that (though still horrible). HTML for kindle.

I had exactly the same experience with my Nest thermostat. I think Nest's assertions about furnace compatibility are a marketing wash. The three support people I spoke to were explicitly disinterested in knowing the make and model of my furnace.

I appreciate the company's consumer electronics mindset, but they need to take the business of running HVAC systems more seriously.

My "dumb" thermostat is back on the wall and successfully running my heating system (to the schedule I've programmed it) as it's done for years.