Vishnuv retweeted
tbh i never quite liked the terminology of “software factory”
but this has got to be the best write up i’ve seen so far that shared all the inner details of how to DIY and how each bottleneck was solved
great reference for anyone exploring custom harness/factories!
New blog post out! Learn how I designed and built my software factory. It covers the architecture, design choices, and important learning along the way. Big shout out to @kunchenguid who inspired some design choices (the daemon) along the way!
open.substack.com/pub/thebui…
Vishnuv retweeted
At @SpaceXSI, Lauren Tan (@poteto) and Roman Ugarte (@romanugarte_) are on the core team that took Grok Bot from first line of code to launch in seven weeks.
It lets you spin up AI teammates with their own identities and logins, and they just keep working in the background. Their marketing team gave a bot its own ad budget. Lauren built an entire feature with her bot in one Slack thread without ever opening Cursor.
They joined us on the @LightconePod to talk about what it’s actually like working next to bots every day, and how the hard part is shifting from doing the work to figuring out what to do and why.
03:36 — Giving Bots Real Jobs
06:49 — One Bot or a Team of Specialists
10:39 — Sharing Bots Across the Company
14:46 — Identity and Permissions for AI Coworkers
18:37 — Bot-to-Bot Collaboration and Trust
22:35 — The Workflow Behind 4,000 PRs a Month
26:00 — Teaching Agents to Write Code You Trust
29:01 — Building a Personal Software Factory
33:44 — The Future of AI Coworkers
36:47 — Rethinking the AI-Native Company
40:13 — Teaching Agents to Check Their Own Work
43:56 — How Much Code Review Is Enough?
47:33 — Codebase Guardrails That Save Tokens
52:20 — When Product Judgment Becomes the Bottleneck
56:29 — From Finance to AI Engineering
Vishnuv retweeted
OpenOrange vs celld
34.8x lower S3 costs for agent workloads
This is made possible by building OpenOrange on Rivet's storage engine. We've spent years scaling the actor model, and Rivet's storage engine was built for this.
celld's architecture balloons costs for:
- Agents
- Workflows
- Background jobs
- Session stores
~
Why?
First, a primer on running databases on S3.
S3 is cheap for storing data. It's about 10x cheaper than AWS EBS replicated across 3 availability zones:
$0.023 vs $0.24 per GB per month
But it gets expensive once you actually operate on that data. Writing one object every second for a month costs:
AWS EBS ~$0
S3 $13
This is because S3 charges per operation, while AWS EBS charges for how much you store and for how long:
PUT costs $5 per million
LIST costs $5 per million
COPY costs $5 per million
GET costs $0.40 per million
On AWS EBS, these operations are effectively free.
The catch is that the number of operations scales with how many objects you have. Every object has to be written with its own PUT, so 1,000 objects cost 1,000 PUTs. Pack the same data into one object and it costs one.
~
So why is celld so expensive?
celld stores each Durable Object as its own S3 objects. So the more Durable Objects you create, the more objects you have, and the more you pay.
Every new Durable Object costs PUTs to its own objects: claiming it, then saving it when it goes to sleep. That's ~4 PUTs and 1 GET per Durable Object in our benchmark.
If you have a fixed number of Durable Objects, this cost is reasonable. But if you're constantly creating new ones, your cost grows with every one.
And agents tend to be high churn. Some agents work for days, but most do a quick, discrete task and never get used again.
These types of workloads suffer, since they frequently create new Durable Objects:
- Agents
- Workflows
- Background jobs
- Multiplayer game rooms
- Chat rooms
- Per-user sessions
- Rate limiting
~
So how does OpenOrange stay so cheap?
Rivet's storage engine is built on an LSM tree (log-structured merge tree). It's the same idea behind mature S3-backed databases like ClickHouse, used to keep performance high and cost low.
Instead of giving each Durable Object (called Actors on Rivet) its own S3 objects, Rivet's LSM packs many of them into shared S3 objects.
Here's how PUTs compare per 1,000 new Durable Objects:
- OpenOrange ~90
- celld ~4,000
When you create a Durable Object or write to SQLite on OpenOrange:
1. The write is replicated to SSDs across 3 availability zones (single-digit ms)
2. After about a minute, writes from many Durable Objects are merged into one batch
3. That batch is written to S3 as one object
~
These benchmarks are modeled on real-world agent workloads, measured at 8 to 25,000 concurrent agents. Each agent lives 60 seconds and writes a turn every 10 seconds. We only measured S3 cost, and both ran on identical hardware.
~
OpenOrange is open source today.
We've attached the benchmarks below.
Vishnuv retweeted
drawn in JavaScript by Claude Fable 5.2
Here's my prompt for this drawing:
> Analyze [/Users/kevin/Documents/art] for hands and self-portraits.
>
> I want you to choose up to 3 to 5 dry art tools (the best combination), or make up your own tools and colors.
>
> It must feel very HUMAN-DRAWN, not digital at all. Not like an SVG. u can spin up subagents but they all must be the same llm model as u. also try not to use headless chrome browser cuz it hurts my laptops memory
>
> I want you to create one masterpiece artwork, drawn with Javascript only. 16:9. Very detailed. Put your heart and soul and energy into it.
>
> It needs to evoke these human emotions, but DO NOT actually have anything from this. dont draw ppl:
>
> i want you to imagine: you drew this art piece and then ur gonna present it to [investor] to raise money. it's the day of the pitch but you don't have a deck. so you show up and its just [name of investor] (cant spell his name), and hes sitting there and he doesnt say anything. you get out ur ipad and you show him the art. he reaches into his desk drawer to grab a check and writes it to you for $1m seed. he congratulates you. then he says "i wanna show you smth". so you follow him into a room and its a karoake room, then he turns on apt by bruno mars and rose and you just sing with him while drinking peach soju. then he shows u around the [investor's] office and u meet others like the other guy. and then he takes you to a room, it's all gold with a giant x hanging from the ceiling. underneath the x is a laptop . he says "post it". so u walk up to the laptop and sign into ur x account. u upload the image with caption "Art drawn with javascript, by claude fable 5.2". then you post it and hte room shakes a bit. theres like 100 comments in the first few mins. suddenly [famous person] retweets ur post. [investor] and the other guy walks u out to the front door and they shake ur hand. on ur way home u stop by in n out burger and order a double double with fries. there's no line at all
>
> ===
>
> Please do your best to capture the emotions of this. it should be absolutely beautiful and use every pixel of the canvas . again , do not have anything from the story in this art. aim for 90s anime look (see oav folder), but not really 90s anime, like something original. just kinda capture the look and feel but its super creative like mind blowing creative really. i like the colors of whats considered 90s anime tho
>
> ok wait
>
> in addition to all of the emtions above, i want u to capture these emotions too (do NOT have any of this stuff in the art pls):
>
> ===
>
> its ur 5th or 6th bday forgot and ur parents just bought u a n64. u plug it into ur crt tv and put in mario 64. the screen shows up and ur stretching marios face for first time. its aamazing. u play for hours and hours and hours. nostalgic . ok now ur parents tell u u gotta stop playin and eat dinner cuz they got pizza hut . the pizza hut guy rings the door bell. he drew an elephant on the lid cuz he knows thats how u always order it. then u eat the pizza but ur hands r still greasy but u go back to playing the n64 before washing ur hands cuz u cant wait .
>
> ===
>
> alright i hope that all made sense, good luck
Skills I use all the time to make insane demos
1. github.com/achimala/dream-lo… - dream-loop, 3D visuals with Blender, image gen and a critic agent
2. hairline.lucasmarkes.com - Hairline, isometric line drawings that answer the pointer
3. emilkowal.ski/skill - Emil Kowalski's design engineering skills
4. github.com/mengto/skills - my own skills for designers and builders
5. rea.tools - REA, reverse engineer how any app or site works
6. fil.design/skills/html-demo - HTML Demo, your app working live in HTML
Vishnuv retweeted
I asked Claude Fable 5.2 to push this further and paint me a masterpiece with JavaScript.
It painted 20 scene frames with photo-real brush strokes along one road from dusk to daylight.
No image model touched this. Oil on linen, made entirely in code.
Prompt in the reply below.
Vishnuv retweeted
You're the reason your mornings are frantic:
You have a thousand dollar phone with a clock on it. Set it. When it goes off, get out of bed.
Spend 10 minutes the night before - take your shower, lay your clothes out in the order you'll put them on.
Then the whole morning routine takes five minutes and you have 40 left. Sit in a dark room and name 10 things you're truly grateful for.
Most people have just never felt the power of being in control of the small things.
DJ Shipley on @hubermanlab
Vishnuv retweeted
i found "executive mindset" to be the biggest differentiator from people who can use agents super effectively
it comes natural for people who had been an executive of a large org before, but it's counter-intuitive for people who haven't done so
a few examples
1. after you send a prompt, do you stare at the chat to see what the agent is doing?
most agent UX today leads people to do that, but a good executive never watches over an employee's shoulder after they've sent an ask - they are immediately on to the next problem
2. do you rely on chat for important decision making?
"just chat" sounds elegant, but there's a reason company executives rarely rely on plain talking for high stake decisions
instead, they ask the team for documents or presentations. it puts a bit more prep work on the team, but the discussions end up being a lot more efficient
3. when your agent did something wrong, do you curse at it and twist its arms to do it right?
some successful executives actually do that as well, but they also do something else afterwards - figure out what allowed the mistake to happen in the first place, and stop it at its root
4. do you spend all your time carefully examining everything to make sure the code is bug free?
now this is getting spicy, but you can pick any popular product on earth and you won't find a single one that's bug free
good executives accept the calculated risk and invest time in the right workflows to catch bugs early before shipping, get alerts if they happen, and making sure fixes can go out quickly - never make themselves the bottleneck by being the tester
our tools, skills, workflows etc change all the time, but building the right mindset is durable and can help you for a long time
if you aren't clear how to acquire the mindset yet, i recommend starting with asking yourself this when you try to decide to do something or not -
"if i have 100x more agents doing 100x more work for me every day, how would i do this?"
do this enough times, and you'll find yourself actually able to control 100x more agents effectively
Vishnuv retweeted
we now live in a time where you can close your laptop and your AI agents keep working
let me explain…
→ Simular just shipped Box: one cloud computer every AI you use can share
→ Grok can start the job, ChatGPT can carry it on, and Claude can finish it
→ log in once… every agent after stays signed in
→ access to desktop apps, Chrome, terminal, files, and downloads
→ jump into live view any time, to even approve a payment
→ it also works with Codex, Claude Code, Gemini, Cursor, VS Code, Gemini CLI, OpenCode, OpenClaw, Hermes, DeepSeek and any MCP client
pretty epic bros. our laptop lids and batteries have stopped being the off switch.
paste this into your agent to get set up:
"Set up Box from box.simular.ai/install"
🤝 Paid partnership
2,000,000,000 PCs sit on people's desks today.
The next billion computers?
They’ll be autonomous computers, driven by agents.
Introducing box by @Simular: one computer for ChatGPT, Claude, Codex, Gemini, DeepSeek and all your other agents.
With box, agents open apps, run the terminal, fill in forms, and stay signed in for whatever tasks come next.
Best places to find breakout startups hiring before everyone else:
• Lenny's top 50 hottest startups: lennysjobs.com/next50
• 92 breakout startups picked by investors: breakoutlist.com
• Weekly roundup of early startups hiring: nextplayso.substack.com
Vishnuv retweeted
New skill: /learn-taste
Collects every design correction you made during a session and turns each one into a strict rule with reasoning, so that the next agent doesn’t make the same mistake.
A great time saver if correct your agent often.
aiforui.dev/skills/learn-tas…
Sometimes I see people calling me a grifter, saying I’m just trying to squeeze as much money as possible from my audience.
And honestly it hurts to read because it couldn’t be further from what motivates me.
We’re in the middle of an AI gold rush. Startups are raising millions, and if you have an audience in tech right now, there’s an insane amount of money to be made.
People don’t realize how many offers I turn down from companies willing to pay $5K for a single tweet promoting their product.
I could probably make a ridiculous amount of money doing this. But I’ve always refused because I don’t want my opinions to be for sale.
And I’ve seen firsthand how much brands are willing to spend on visibility with my HYROX sponsorship experiment. If money was my goal, I’d just double down on that.
The last few years have completely changed my life. I’m financially free, I have everything I need, and I genuinely don’t care about making more money just for the sake of it.
Don’t get me wrong, I LOVE making money from my apps.
Not because I need more of it, but because someone paying for something I built is the ultimate validation. It means I’ve created something valuable enough that someone is willing to pull out their credit card.
And if one person is willing to pay for it, maybe others will too. That’s what makes building businesses so exciting to me.
But what actually gets me out of bed is making weird stuff, experimenting with ideas, and seeing people use things I’ve built.
From 2015 until 2022, I was building apps alone, with barely any entrepreneur friends. I felt lonely.
Now I’ve finally found my people, made enough money to live comfortably, and get to do what I love every day.
I don’t need more. I just want to have fun building stuff with people who love it as much as I do. ❤️
This. This man right here is pushing out real stuff!!❤️
Working with agents across multiple repos is a pain
Where do you put the skills? The steering files? The tooling?
Let me propose a solution I've been using for months (that I thought was too boring and obvious to post)
It's a pattern called the Home Repo. Enjoy:
16
This is going to be such a game changer for microservices if it is what I think it is. I was trying something similar but mine was bunch of slop which i built in haste but can't wait to try this pattern. Time for some refresh😌
Published a proper article on the Home Repo pattern - the simple solution to running agents across multiple repos.
Contains a prompt you can run on your org's repos to help you set it up for yourself.
Enjoy your new home
aihero.dev/the-home-repo-pat…
1
49
Vishnuv retweeted
I've been going through @emilkowalski's aiforui.dev course and I have to say it has probably been the best investment I have made this year.
Thank you for putting together such a simple to understand and easy to apply course that is both deep while being super practical across different domains...
Vishnuv retweeted
Move over, B2B and B2C - the B2A economy is here
Agents are the primary users of the Internet, and soon will be the primary customers for every product
Our thesis @a16z on the massive opportunity here 👇 (cc @venturetwins)
Check out my new grok bot: teapawt.net/var_wish/status/212171…
Yes. I made a Grok Bot template for this: Shoebox. It sorts years of photos into year and trip folders, finds exact and near-duplicates even with different file names, and shows you the plan first. It never deletes; dupes go to a review folder. x.ai/bot/oA2LC3aRpxUKKBX03cy…
23

















































