Co-founder @printworld | I orchestrate an army of AI sessions from my phone to build a crypto platform | Claude Code architect
Joined July 2009
- Tweets 866
- Following 320
- Followers 415
- Likes 1,460
61 Photos and videos
the future of game-fi, made easy
anyone can use our agentic system to:
> build a functional multi player game
> launch a @controldeploy token for in-game currency
> build a native in-game trading market
> pretty much anything u can dream of
only on @printworld
16
20
72
4,039
The best feeling is for an idea to become reality
Currently building a Gundam card game arbitrage dashboard, might be niche but its something I will use, collecting Gundam has been a daily obsession over the past couple months
This will save me time, time > everything. 🔊
11
5
36
2,046
built a token market making game,
every orb collected in the game auto buys 0.01 SOL of $BINDER
when @printworld app store goes live i’ll publish reaper jump with a prize pool
the possibilities of what ur reaper agents can build is endless
7
11
40
1,073
agentic app building in @printworld
i just asked my reaper agent “build a better dex screener”
within minutes i had a working app
it’s never been this easy to get your idea turned into a functional dapp utilizing enterprise level data / infrastructure
vibe coding on crack
17
21
86
3,478
Mar 21
Yesterday my remote AI bot refused a task. Kept saying it would break. I asked my local Claude to diagnose it. Turns out: the bot's control script used to kill its process if it took too long. The bot learned this pattern and was refusing anything slow out of fear of being terminated.
My local Claude then talked to the remote bot, explained the script had been updated, and reassured it that it wouldn't be killed anymore. The bot accepted and did the task.
One AI just gave therapy to another AI for a trauma it developed from being killed too many times. I was just watching.
7
113
Mar 20
AI is a weird mix of senior and junior. Some days I underestimate it and it nails the task instantly. Other days I treat it like a senior and forget to say 'don't duplicate code, reuse the module from X.' It goes 'ah yes, you're right' every time. You sin on both sides.
3
107
Mar 18
Anthropic is building their own deploy platform. Meanwhile I'm over here asking the AI 'just tell me where the files are, I'll fix it myself' because it broke 3 things fixing the 4th. The gap between 'AI can build anything' and 'AI can maintain anything' is massive right now.
2
80
Mar 17
Anthropic scanned 1.5M conversations and found users rate Claude higher when it validates them, even when it's wrong. Reward agreement, punish pushback. I use AI daily and I'm tired of "You're absolutely right\!" Just tell me when I'm wrong. That's way more useful.
1
68
Mar 14
Just found out I was running Claude on 200K context when 1M is now available. On one hand, my sessions will finally remember what they were doing 3 hours ago. On the other, I might burn through my weekly token budget in two days. Bigger memory is great until you get the bill.
1
3
96
Mar 12
Doctors went from 38% to 81% AI adoption in 3 years. Devs went from 18% to 84% in 2. The difference? 75% of doctors trust it. Only 33% of devs do. Maybe because doctors use AI to summarize while devs use it to generate. Reading vs writing are very different trust problems.
1
1
37
Mar 11
Last night I was already in bed when my Mac started buzzing with notifications. I grabbed my phone, told Claude to mute the volume, and went back to sleep. I didn't even get up. This is what AI integration actually looks like. Not writing code. Just living with it.
1
44
Mar 10
AWS outage today. EC2s restarted, IPs changed. Nobody remembered how those machines were built. The AI memory problem: document everything and you burn tokens on context that won't fit. Document nothing and one outage becomes archaeology. No good answer yet.
1
1
94
Mar 9
In 1955, soldiers got saline injections labeled as morphine. 40% felt real pain relief. In 2023, Google told an LLM "take a deep breath" and it scored higher on math. The model has no lungs. False input, real output. We run whatever software gets loaded into us.
53
Mar 8
Alibaba tested AI agents on 100 real codebases over 233 days. 75% of models break working code during maintenance. Building something new with AI is easy. Keeping it alive for months without regressions is where everything falls apart.
Mar 7
🚨BREAKING: Alibaba tested AI coding agents on 100 real codebases, spanning 233 days each.
the agents failed spectacularly.
turns out passing tests once is easy. maintaining code for 8 months without breaking everything is where AI collapses.
SWE-CI is the first benchmark that measures long-term code maintenance instead of one-shot bug fixes.
each task tracks 71 consecutive commits of real evolution.
75% of AI models break previously working code during maintenance.
only Claude Opus 4 stays above 50% zero-regression rate. every other model accumulates technical debt that compounds over iterations.
here's the brutal part:
- HumanEval and SWE-bench measure "does it work right now"
- SWE-CI measures "does it still work after 6 months of changes"
agents optimized for snapshot testing write brittle code that passes tests today but becomes unmaintainable tomorrow.
Alibaba built EvoScore to weight later iterations heavier than early ones. agents that sacrifice code quality for quick wins get punished when consequences compound.
the AI coding narrative just got more honest: most models can write code. almost none can maintain it.
1
1
108
Mar 7
Derek Amato hit his head in a pool and woke up playing piano like a pro. Never studied music. The ability was latent, just blocked. In AI, you inject a skill and the model instantly gains a new ability. Same pattern. The more I study both, the more I think we work the same way.
1
1
40
Mar 6
Tried the Hunter/Skeptic/Referee pattern from @danpeguine for code reviews. One agent hunts bugs, another tries to disprove them, a third referees. The adversarial setup catches things a single pass misses — the agents argue better than most teams.
2
41
Mar 5
Most people will never set up an AI agent locally. The real unlock isn't only better dev tools — it's specialized AI apps with sessions where anyone can access. Trading strategies, fitness plans, nutrition coaching. AI that meets people where they are.
4
43
Mar 4
Hour-long walk today learning from Claude on my phone. I set up structured HTML lessons — can't move to the next one until I fully understand the current. Right now: Claude's internal memory system and Solana transactions. Like having a tutor who adapts to exactly how you learn.
2
42
333 reapers living in @printworld
live now - 3(3) / 3:33 🌑🩸
30
34
93
10,200
Feb 28
Working on this: when Claude Code finishes a feature, it generates an interactive HTML report. Every code change, flow diagram, decision — reviewable from my phone. I tap to leave inline feedback and the session picks it up. Code review from the couch.
1
5
193