MB (@0xdef1) header
MB (@0xdef1) profile picture

Sawtooth iOS beta: https://xitter.gq/t.co/cubbmzX6yR

Joined January 2021
Photo by MB (@0xdef1)Photo by MB (@0xdef1)Photo by MB (@0xdef1)Photo by MB (@0xdef1)Photo by MB (@0xdef1)Photo by MB (@0xdef1)Photo by MB (@0xdef1)Photo by MB (@0xdef1)Photo by MB (@0xdef1)Photo by MB (@0xdef1)
Pinned Tweet
I ran a mobile studio for 7 years I built this app in the last 3 weeks with a few agents, from my kitchen counter, mostly while doing other things. IMO it's as good as anything we ever shipped It's a live music discovery app. Beta link below 👇
11
2
34
1,664
No, stop with this bullshit
This quoted post is unavailable.
1
15
568
When I coded for a living, I felt like a slacker. ~10% of my time was spent typing in an IDE Years later, I realized pacing around my office daydreaming about code was the real job Agents do the typing now. I still pace and daydream. It's still the most valuable thing I do
9
40
1,018
With agents, I am reminded of something I used to tell my team all the time when I ran a studio: “Know when to ask and when to tell” You can’t be a wimp and always ask clients what they want. Our POV is why they hired us But you also can’t dictate everything. You don’t know all the hidden complexity or constraints they have
2
22
788
I took a year off, then spent two months working exclusively with agents I just started a new job and am now working with people again and realized something: Humans are slow as fuck bro
9
51
1,532
It would be nice if the bible had clarified if adultery is worse than decreasing shareholder value
28
168
4
2,106
43,071
i don't know shit about nvidia or jensen but i also don't think extramarital affairs should displace the ceo of america's most successful and important company either something else is afoot or it's terrible decision making, I can't tell which
4
11
439
Frontier models' tendency to think ahead and plan for edge cases is valuable for most coding tasks, but imo harmful for abstraction and data modeling It frequently builds flat, unnormalized relations , and I need to ask for diagrams and tell it YAGNI / DRY / simplify
10
15
1,041
Favorite ai thing of the day: “create an agenda for this session and keep it updated” I kept losing track of small side tasks that would come up, especially when an agent responded with multiple questions in a single turn The agenda keeps track of all those loose ends It can be short lived just for a session so it doesn’t have to pollute your harness or other session context I’m sure this could be a skill with some slick UX but a simple prompt works pretty fuckin good
12
18
656
"show me with an ascii diagram" is my favorite prompt of the day turns walls of text into a much more easily digestible thing doesn't require a fancy web-based custom harness and doesn't create artifacts / side effects
12
16
890
Lots of people are dunking on this but I think a subtlety is lost Slop is bad in both, but humans are far better at recognizing (and therefore more likely to fix) frontend slop So vibe coded front ends end up ok, but only because the human is better at frontend slop detection Frontend devs know this — bugs get reported to the frontend devs first, they end up triaging everything
one of the most underrated asymmetries right now is that slop is totally fine for the backend and basically fatal in the frontend. if you’re backend is slop, the app still works. if your front end and overall experience is slop, the *product* does not.
7
19
1,217
"reading fatigue" is real. models produce so much text that I spend most of my day reading, and very little time deciding or acting I've tried things like "terse mode" or "caveman mode," but still find i'm fried from reading halfway through the day
6
1
17
995
I've started to notice that agents, even with the same boot docs, change personality by project Backend agent: rigid, assertive, pedantic. iOS agent: creative but careless Is this an emergent property of the training data? Or me projecting my own biases? Who can say...
8
14
826
Waking up to agents that did work for me over the night is still something that amazes me I’m sure at some point we’ll all take this for granted but let’s just reflect for a moment on how incredible this is 🙏🙏🙏
7
1
20
2,521
wtf do you tell young technical people to learn today my hunch is learning to code will be valuable the way learning assembly was valuable after high level languages existed. helps a little but not required CS still valuable, but curriculum needs to zoom out?
Replying to @master_jpma
I actually don't know I know that having a ton of experience gives me a leg up in creating robust software with agents But I also know that the way I gained that experience (hand-coding for ~30 years) would be a complete waste of time
11
19
1,304
What are you using Fable for right now? For literally everything I’m doing, Opus 5.5 is both better and cheaper Opus also beats the OpenAI models in my limited experience
7
10
1,203
Vibe coding is used as a pejorative, and I get why, but it's just another name for something we've always done We just used to call it "quick and dirty" or "prototyping" or "shitty mvp" Agents operate that way if you prompt them that way, but they're more capable than that
vibe coding is easy but how do you maintain a 70k line codebase?
20
1
26
1,779
North Korea's hackers dumped $750M of stolen crypto on DEXs, but it cost them $107M. MEV bots & Traders made over $50M. ETH hit $3,135 on-chain and AMM pools were temporarily emptied for their reserves. So what exactly happened to the liquidity, and where did the $107M go?
5
1
3
22
1,776