on a journey to build great things with a box of scraps building things https://xitter.gq/t.co/GdPO0ANoLp and https://xitter.gq/t.co/GD9jdjCqL7 local retard in motion
PHL
Joined February 2022
- Tweets1.6K
- Following712
- Followers970
- Likes7.8K
if you're on the fence on whether you should build verification skills; just do it
work on the following:
- empirical testing (e2e and feature journeys)
- formal verification
- fuzzing, static analysis
do the above and even a weaker model should have a harder time making mistakes
46
day 5! i'm just logging whatever I'm working on atp
But I finally set up my trading engine - here's what that looks like:
1. I have watched "sources" based on a "market"
2. Does source produce an "event delta" that reprices that market?
Yes? → BUY
No? → NO BUY, log
I've been working on building this for so long now, and with the help of finally locking-in™, I've finally gotten to a point where it's up and running.
In systems like this the most important thing is actually the data that you gather, so for now...just gather as much data as possible. Phew.
Phew OK. Time to lock in even further. 😈
Day 4 of building
Spent today recalibrating my dot and setting up routines together.
I have a feeling that spamming your thoughts over call with an agent is going to be the meta, as it really allows you to think more collaboratively over writing.
Its a lot more responsive too - if your personal agent cuts you off a lot, calls were engineered specifically to **not** have that issue.
Tonight we set up a daily close-out to review what happened, capture what’s pending, and plan tomorrow.
This is what I think personal assistants should become: something that learns how you work and helps you make sense of your life over time.
Dots are the closest I’ve felt to having a real-life Jarvis.
But also - look at how cute he is!
26
just when I thought t3 code couldn't get any better.
you can connect your external VM/macmini through SSH and actually access the browser from that session!
this is a complete life changer, especially after codex completely gutted the ssh experience.
life changer really.
2
115
Day 4 of building
Spent today recalibrating my dot and setting up routines together.
I have a feeling that spamming your thoughts over call with an agent is going to be the meta, as it really allows you to think more collaboratively over writing.
Its a lot more responsive too - if your personal agent cuts you off a lot, calls were engineered specifically to **not** have that issue.
Tonight we set up a daily close-out to review what happened, capture what’s pending, and plan tomorrow.
This is what I think personal assistants should become: something that learns how you work and helps you make sense of your life over time.
Dots are the closest I’ve felt to having a real-life Jarvis.
But also - look at how cute he is!
Day 3 of building!
My CI keeps getting throttled because my VM keeps running out of storage.
I have 3 runners on it. The first thing I tried was a daily cleanup at around 4 AM to prune unused files, but storage is still a problem because it only pruned up to ~40GB at a time. Which is nowhere near enough to handle the demand during the day.
Now I’m asking Codex to see if all three runners can share a cache instead of keeping their own.
Let’s see if that helps.
I really like Dots from @OpenAI
Almost switched to Grok Bot, but after deciding to give jarvis (my dot) another try, he's actually doing really well now. Performance is comparable, if not better.
Copying what works from the Grok bot marketplace is making a lot of difference!
I've experimented by importing a lot of the important context from dr Eggbot and other GrokBots.
1
2
88
Day 3 of building!
My CI keeps getting throttled because my VM keeps running out of storage.
I have 3 runners on it. The first thing I tried was a daily cleanup at around 4 AM to prune unused files, but storage is still a problem because it only pruned up to ~40GB at a time. Which is nowhere near enough to handle the demand during the day.
Now I’m asking Codex to see if all three runners can share a cache instead of keeping their own.
Let’s see if that helps.
day 2 of building my own shit
CoreMVP is starting to show up on google
2.99k impressions and 13 clicks in the last 3 months 💀
impressions are picking up, but now i need to figure out how to turn those into actual visits
starting by looking at what people are searching for and making those pages more useful
i'm already working on a /help page which should capture most of the beginner queries beginners use!
let’s see where this goes
why is it everyone's last day at anthropic
Friday was my last day at Anthropic.
Two years ago, when Claude Code was a small internal tool, I asked it to build an RL env. I steered it for tens of hours but never manually wrote a line of code. It worked, I was hooked, and I joined the team.
When we first launched, most people approved every permission prompt by hand and mostly used Claude to answer questions and write tests.
Today, the default is trusting Claude's judgment to auto-approve. Claude does large migrations and builds new products.
Helping build Claude Code into an industry-defining product has been the journey of a lifetime.
Thank you to the incredible team behind Claude Code, Cowork, and Tag. These products are the sum of thousands of great ideas from you. I'll miss working with you all.
Thank you to our users for embracing a new way of working. Your constant feedback shaped so much of what we shipped and your stories made the work so meaningful.
Excited to see what you all make next!
57
a few bets I'm making in life:
- finding a good wife
- my most important investment is myself
- all financial investments come from systems
- need to be surrounded by the most talented people possible
- I need to a/b test everyday for the rest of my life
what am I missing
36
dont fall for the propaganda
this is only possible with infinite compute to fix the spaghetti code of previous iterations
99% of us won't get to experience this euphoria. not even on a 200usd plan
this is basically the bitter lesson at work
the smartest models will need fewer and fewer tokens, they won't need to verify their work or check it for bugs, they will just be right
31
no matter what I will not fall victim to the AI overlords who pushes 2000 PRs a month
i still care about a clean codebase i can understand
what about you?
1
27
day 2 of building my own shit
CoreMVP is starting to show up on google
2.99k impressions and 13 clicks in the last 3 months 💀
impressions are picking up, but now i need to figure out how to turn those into actual visits
starting by looking at what people are searching for and making those pages more useful
i'm already working on a /help page which should capture most of the beginner queries beginners use!
let’s see where this goes
indie hacking is dead, apparently
i think the bar just got higher
so i'm putting that belief to the test, starting with CoreMVP
i'll share what i decide to build, how i try to get it in front of people, and what happens after that. the attempts that go nowhere belong in the story too
if you're building your own thing, i want you to leave each post with something you can try—not just another screenshot of someone else's progress
there'll be product decisions, experiments and probably a few questionable ideas along the way
day 1 of building my own shit
come steal the useful bits
i want CoreMVP to be an option for builders who can code and want to sell their first app, but don't want to spend hundreds of dollars on a starting point
i'm building from the philippines. paying $200 or $300 for a template is a big decision for a lot of people here
ShipFast's starter is $199 and MakerKit's Supabase Pro is $299 right now. they're different products, so compare what's included. free starters exist too
CoreMVP Solo is normally $99 one-time. eligible first-time buyers can currently get it for $49 before tax with FOUNDING20
but price is only part of it
there's already a lot to think about when you're vibe coding. giving your agent a good starting point means it has less to invent. small decisions around security, access and billing can turn into much bigger headaches later
CoreMVP includes 15 agent skills and four procedures for planning, challenging, implementing and reviewing changes, with guidance for Codex and Claude Code
billing is one of the parts we've spent a lot of time getting right, and we'll keep improving it. we support BOTH Stripe and Polar for one-time payments and subscriptions
that includes checking webhook signatures and handling repeated events. if a provider sends the same payment notification again, your app needs to recognize it rather than blindly process it twice. that's the kind of idempotency detail that's easy to overlook when you're focused on getting the app working
payment access also comes from stored, verified billing state. landing on a checkout success page isn't proof that someone paid
you still need to configure your services, test your payment flows and build your product. hosting and provider fees are separate. a template doesn't make those responsibilities disappear
you get the source for one developer, commercial and client-project use, and lifetime updates
if the smaller feature set fits what you're building, Hikari is free and MIT licensed. if you're still learning JavaScript, start with free learning resources. you don't need to buy something from me to start
$49 is still money. look at the docs and demo first
FOUNDING20 ends november 9 at 12:00 am manila time, or when its 20 uses across Solo and Team are taken. checkout confirms eligibility and the final total
4
1
128
