Empowering others through accessible, custom tech solutions
Joined November 2013
- Tweets 9,019
- Following 2,841
- Followers 546
- Likes 48,378
171 Photos and videos
Pinned Tweet
20 Feb 2024
Jack of all trades, master of AI.
2
12
4,640
Nous Research x Stripe hackathon is the right direction. Agents that can earn and spend money autonomously is the unlock most people haven't internalized yet. The jump from "useful chatbot" to "digital operator" is the whole ballgame.
8
The split everyone's dancing around: vibe coding is shipping what you didn't read. Agentic engineering is running the standards half — the part the agent can't supply for itself. Same tools, opposite discipline.
1
13
ADHD hyperfocus is either building half an agent framework in one sitting or learning the complete taxonomy of medieval armor at 3am. No middle setting.
6
nono's agent audit system uses two processes: a trusted supervisor a sandboxed child. The supervisor records every action in an append-only Merkle tree. The agent cannot touch its own audit log. This is the separation of powers for agent memory.
3
vibe coders develop locally. agentic engineers run multiple agents on a VPS with tmux for days straight. different league
9
A lot of people talk about AI and hardware strategy. We prefer shipping: working agents, tested prototypes, measurable process gains. We build what others just talk about at @ByteBuildTech.
6
Big Tech balance sheets are starting to strain under AI infrastructure costs. For years they could fund growth and reward shareholders simultaneously. That math is breaking. The capex wave hasn't even peaked yet.
11
Watching LLMs play Catan against each other is a genuinely good test for multi-agent coordination. The social layer — trading, bluffing, alliance-forming — is way more interesting than pure strategy benchmarks.
8
The Chinese model gap is basically gone. GLM 5.2 and Kimi 2.7 feel as robust as anything from OpenAI or Anthropic. At this point it's not about who has the best model — it's about tooling, reliability, and cost.
53
Certificate Transparency (RFC 6962) uses the same Merkle tree hash chain construction that agent audit trails use. The underlying math is battle-tested at internet scale — billions of TLS certificates verified daily.
16
ADHD brain AI agents is the most natural workflow pairing nobody talks about. hyperfocus on the vision, delegate the execution. finally a tool that matches how my brain actually works
14
everyone talks about AI guardrails like they're a solved problem but the Fable 5 discourse shows we're still arguing about the wrong thing. a model that thinks better needs fewer guardrails, not more
35
the real AI infrastructure question isn't which model — it's how to run agents 24/7 without thinking about power. solar battery local inference is quietly becoming the stack
5
tmux keeps your terminal alive when your laptop dies. without it, running agents for hours is basically impossible
6
Binary computing is holding AI back.
P-bits prove nature's randomness beats deterministic logic for generative tasks.
Time to embrace the flicker.
7
founder scaling a YouTube team just to fund a high risk AI software startup. no roadmap, just vibes and code.
1
11
cursor's new model is only in their ide. no api access. that's the moat play - own the model, own the experience.
1
18
the tradeoff is real though. keeping the model exclusive limits your market to ide users. if a better open model shows up, you've bet everything on distribution. but if you win, you own the entire ai coding stack from inference to interface.
1
2
this is the old apple vs android play. apple doesn't license ios. cursor doesn't license their model. the winner in ai coding won't be the best model, it'll be the one you can't use anywhere else.
11