Hanako

@hanakoxbt • 23,010 subscribers

Building my name | ai researcher | dm open

Shorts

there are four types of agent loops. most people only know one. loop engineering is a choice between four structures, each handing off one more job than the last. every one answers two questions: what starts a run, and what ends it. hand-run, you answer both yourself, every time. 1) turn-based → you prompt, it acts, you review, you prompt again. both jobs stay with you. use when requirements are still forming. 2) goal-based → "/goal hit Lighthouse 90, stop after 5 tries." an evaluator checks, a no sends it back. use when the outcome is measurable but the path isn't. 3) time-based → a clock fires, it runs "check the PR, fix CI," then waits. /loop local, /schedule survives a closed laptop. use for recurring work. 4) proactive → no human present. it watches a channel, spawns triage, fix, and a reviewer, closes the task itself. use for standing duties you can't predict. not which one is most advanced. whether your task is exploratory, measurable, recurring, or standing. the more you hand off, the less you babysit. full breakdown in the article below.

478,460 görüntüleme

If it takes 40 slides to explain your brand’s personality, I have bad news. I wanted to see whether three words could carry a feeling on their own. So I made this type loop with Pexo from Pexo. I wanted it to feel playful without explaining itself. Seeing the words move was how I could judge whether that feeling was actually there.

72,907 görüntüleme

this is certified f*cking gold a 23-year-old from China runs 300 Claude Code agents at once with Jev engineering, and not one of them gets to lie to him he opens the dashboard live: 300 Claude Code agents running in parallel Jev checking every output against its source in about 100 milliseconds pointed at 100 companies in the EV market: > pass 1 - 12 rejected: wrong revenue, dead citations, empty fields > pass 2 - 3 rejected > pass 3 - zero. every row filed straight into an Obsidian vault the whole trick is the split Claude Code does the work, Jev decides whether each claim actually matches the source Jev never writes a second opinion. it answers yes or no with a confidence score, at $0.042 per million tokens in and the loop won't stop while anything is still wrong an analyst team bills weeks for 100 verified company profiles. he gets them in one run one model writes, a cheaper one decides, and nothing ships until it says yes

36,798 görüntüleme

an agent is four parts in a loop. you own one. the other three break it. that's why the demo works and prod doesn't. you can't debug what you can't see. 1) the prompt → what you tell the model each turn. you own this one. good. 2) the context window → what it sees right now. the framework fills it with junk, and you never notice until it rots. 3) the tools → what it can do. you own the list, not when or why it fires them. 4) the control flow → what happens next, when to stop. the framework owns this. it's what breaks at 80%. own all four and your agent stops being a magic trick that works on stage and dies on call. this isn't my idea. it's the 12-factor agents guide (24k stars) github: the whole thing every serious builder ends up rewriting their stack around. full breakdown in the article below.

38,387 görüntüleme

Videos

LIVE

1.2k

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Streaming Now

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

HD live stream

Exclusive private shows

1.2k viewers online

Current Status

Live

Private Show

Join now for exclusive access

Free preview available • Premium content

2:02:37

Andrej Karpathy spent 8 years at OpenAI and Tesla Last week he compressed everything he knows into one free 2-hour lecture Prompts → Agents → Harness → Graphs → Loops People spend $15k on bootcamps that teach less than this You probably don't have 2 hours right now Don't let it vanish from your feed Watch it, then read the article below ↓

Hanako

161,557 görüntüleme • 14 gün önce

0:38

a Citadel intern told me something at a party he probably shouldn't have it was on a rooftop in brooklyn. i mentioned i trade prediction markets. he got quiet for a second. "we have a model for that. it scores every contract on four factors. when all four align we enter. when any breaks we exit. that's it" i asked what the four factors are. he looked around. then said it fast like he was confessing. "cross-market divergence. disposition coefficient. capital velocity. pair network correlation" I didn't know what half of that meant. but i memorized it. went home. 11pm. opened Claude. "here are four scoring factors from a quant fund. build a terminal that runs all four on prediction markets" Claude asked one question: "Where's the data?" I sent him one repo: 86 million trades. every wallet. every entry. every outcome three weeks later i'm sitting in my apartment watching a screen i barely understand print money. the disposition meter alone changed everything. it measures how you exit - not how you enter. top wallets capture 86% of winner value and cut losers at 12%. everyone else captures 58% and holds losers to 41%. same exact entries. the exits make it a completely different game. capital velocity: 49x. every dollar gets recycled 49 times before the average trader recycles once. the terminal found 42 pair correlations across 11 markets. when MSFT beats Q3 is priced at 80c but the model reads 93% - it enters. when the gap closes 2 hours later - it exits. no opinions. no news. just four numbers that either align or don't. his fund runs this with a floor of PhDs and $800M AUM. my setup: > Claude - $20/month > VPS - $5/month > poly_data repo - free > Polymarket API - free $25/month. no team. no office. no Bloomberg. 280 trades so far. 70% win rate. $800 seed. four bots splitting the work: pulse_alpha +$299. arb_hunter +$558. trend_rider +$337. cal_engine +$719. +$11,514 total. copytrade here: he texted me last week. "delete everything i told you" too late.

Hanako

1,742,823 görüntüleme • 7 ay önce

37:03

Anthropic Claude Code engineer: "If you're watching Claude write code, you're the QA tester. That's not what you're paid for." In 37 minutes he lays out how to get your keyboard out of the hot path entirely. The shift is /loop. You tell Claude to wake up every 10 minutes and babysit your PRs, and it just does it, while you're nowhere near the laptop. Routines do the same in the cloud, so the work keeps running with your machine closed. He caps it with remote control: any session, on any surface, driven from your phone. Watch the full talk, then grab the setup below.

Hanako

692,778 görüntüleme • 4 ay önce

1:18:01

Sam Altman (CEO of OpenAI): "You don't need to write prompts anymore. You need loops and graphs that write them for you." in 78 minutes he explains how to use LLMs better than almost anyone does Prompts → Agents → Loops → Graphs that is the whole difference between using AI and having AI work for you same model, same tokens, and the only thing that changes is the shape you run it in watch it today, then save the full graph engineering guide below

Hanako

290,293 görüntüleme • 2 ay önce

43:51

Sam Altman (CEO of OpenAI): "Every night I have a few thousand agents running with GPT-6 Astra" a few thousand of them, working overnight, watched from his phone in 43 minutes he shows what one person can build now, and how long that used to take a whole team the part nobody copies is that he is not watching most of them at any given moment watch it today, then save the full breakdown below

Hanako

164,483 görüntüleme • 2 ay önce

28:32

Anthropic engineer: "The agent doesn't remember anything. So we built a second set of agents whose only job is to dream about the first ones." They wait until you log off, then reopen every session you ran, fact-check the first agents, merge the duplicates, and burn anything that went stale. One dreaming agent per session, up to 100 at once, built to miss nothing. 95% of the tokens come back cached, so an entire memory rewrite costs almost zero. This is the part of the workflow nobody is talking about yet. Watch the full workshop, then save the setup below.

Hanako

336,076 görüntüleme • 5 ay önce

0:50

Loops vs. Graphs, clearly explained! loops are great, and they have a ceiling you can watch happen: a loop goes around. it produces, checks, corrects, and goes around again. after six passes you have one job, done very well. after six hundred passes you still have one job, done very well. Graph engineering fixes this by moving the decision up a layer: not how well one job gets done, but which jobs exist to be done at all. you need both, and here is the sentence that resolves the whole confusion: the loop lives inside a node. the graph lives between them. ↳ inside one unit: produce, check, correct, repeat until green ↳ between units: split, fan out, merge, gate, send back Prompts → Context → Harness → Loops → Graphs the loop does not go away when you build a graph. it moves inside, and now there are three of them running at once on three things you would never have thought to run. the trick is being selective about what becomes a node. only spend a model where judgment lives. merging, ranking, deduping and schema checks are edges, and edges are code. free, instant, and they cannot be argued out of a verdict. one thing to know before you scale it. a loop that cannot fail is not a loop, it is a repeat with a bill attached. and the check people write is almost always the wrong kind. ↳ the test suite exits 0 is a check. the diff touches only the files in the plan is a check ↳ the output looks good, the model says it is confident, no errors were raised, none of those are checks that last one catches careful people. absence of an error is not evidence of correctness, and a loop built on it will confidently repeat a mistake until the budget runs out, with a clean log the whole way. and the one that eats whole nights: when a unit fails, return that unit, not the batch. send back four slices because one failed and you have just rewritten three correct ones. do it twice in a run and the run never converges. below i have quoted my full guide on graph engineering. it covers the three topologies, the verifier patterns, and where the gate should actually open. save this and read it below ↓

Hanako

129,105 görüntüleme • 2 ay önce

0:39

an ex-Citadel quant told me a salary is a short position on your own time we were at a dinner in new york. friend of a friend. he asked what i do. "software engineer. $4,200 a month" he said one simple thing "you're mass-selling the most valuable asset you have. your hours. at a fixed price. with no upside. that's a short position on your own life" i didn't have a response. then he showed me his phone. a terminal. live trades. +$39,453 running for 6 weeks. "i built this with Claude Opus 4.7 and one open source dataset. took a weekend. haven't touched it since" i asked how. "gave Opus one repo and one prompt. find who wins on prediction markets. find why. copy their edge" 2,900 stars. 36GB. every trade on Polymarket and Kalshi. every wallet. every resolution. Opus read the whole dataset friday night. by sunday it had a live terminal scanning 1,400 markets per hour. 8 detectors running in parallel. entering when void score exceeds 90. exiting when it drops below 75. no human override. > Senate filibuster. locked at 42c. now 98c. +$1,133. > Neuralink approval. entered 52c. now 98c. +$1,396. > Fed 0% rate. entered 35c. now 97c. +$1,147. > BTC 250K. entered 8c. now 98c. +$1,258. > Megaquake. entered 6c. now 97c. +$1,039. the part that broke something in me: latency arbitrage. Polymarket updates in 19ms. Kalshi in 63ms. the bot trades the 44ms gap between them. automatically. every time. "we had four people at Citadel doing what your bot does with latency alone" copy mirror tracking 6 positions: > Senate fili +$1,252. void signal. 10%. > BTC 250K +$958. zero signal. 2%. > Neuralink +$1,384. sign signal. 9%. > Fed 0% +$863. fill signal. 8%. > Megaquake +$1,070. bridge signal. 18%. > META AI +$979. gas signal. 5%. 972 trades. 83% win rate. kelly f+ 0.101. avg profit $10.65. drawdown -1.4%. maker rebate tracker collecting $203 across four platforms while the bot trades. Polymarket +$114. Kalshi +$40. dYdX +$28. Betfair +$21. $1,800 seed. +$39,453. i quit the next morning. one email. "i'm done. thank you for everything." copytrade setup: texted him a week later. "you ruined a perfectly stable career" he replied "no. i closed your short position"

Hanako

380,375 görüntüleme • 6 ay önce

0:52

ANTHROPIC LEAKED A 6-AGENT SETUP THAT TURNS ONE PROMPT INTO A FINISHED PR your name shows up in this process exactly once - at the very end, for 5 minutes. spec → planner → 3 builders → critic → scribe → PR → back into the spec the planner runs once and the whole line inherits its decisions - a bad plan executed by five perfect agents is still a bad result. 3 lanes build in parallel - and they only work because the planner cut the steps so none of them reads another's output. tests loop until green and red goes back to the coder - the critic never sees a broken build. the critic rejects to the plan, not the code - tests ask does it run, the critic asks should it exist. the scribe writes the PR from the trace, not from memory - cut this seat and you get 8 PRs nobody can tell apart. the back edge into the spec is the whole trick - today's failures become tomorrow's constraints without you typing them. one human step in the whole thing: approve or send back - 5 minutes instead of 5 hours. save this and read the full graph engineering course below ↓

Hanako

164,416 görüntüleme • 2 ay önce

0:40

your agent loop needs 8 exits. most people ship only one. (explained with triggers) 1) goal met → an evaluator scores the output against a rubric, and the run stops on a pass. → fires when the work is measurably done, not when the model says it is done. 2) turn cap → a hard ceiling on iterations, counted and enforced by the harness, not the prompt. → fires on the task it was never going to finish, before you pay to find that out. 3) budget cap → a limit on tokens or dollars, whichever one runs out first. → fires mid-run, which is exactly why it is the exit that saves you the 3am bill. 4) wall clock → a deadline on elapsed time, independent of how much progress was made. → fires when the run collides with a deploy window or the start of business hours. 5) no progress → hash the state every turn and compare it against the last few. → fires when three turns in a row change nothing. busy is not the same as moving. 6) human interrupt → an approval gate before risky steps, plus a kill switch that lives outside the loop. → fires whenever you decide, and it is the one exit the model cannot argue with. 7) error threshold → a counter of consecutive failures that resets on any success. → fires at n in a row, so it halts instead of retrying into the same wall all night. 8) external event → a webhook or a poll on whatever the task was actually about. → fires when the PR merged or the ticket closed and the work stopped mattering. a loop with one exit hangs. a loop with eight is a system. write the exits before you write the prompt.

Hanako

182,236 görüntüleme • 3 ay önce

1:00

ANTHROPIC LEAKED A 4-AGENT SETUP THAT CUTS A CODEBASE AUDIT FROM 3 DAYS TO 20 MINUTES you point it at a repo and walk away - it comes back with what breaks, ranked, patches already tested. repo → map → 4 auditors → rank → fix → verify → report → back into the map the map cuts the repo by blast radius, not by folder - skip it and four agents audit the same three files and miss the one that ships broken. 4 auditors run in parallel with separate contexts - deps, secrets, dead code, hot paths, and none of them sees another's findings. rank is code, not an agent - sort by what breaks production, drop the duplicates, zero tokens. the fixer only opens patches for the top slice - a hundred findings nobody acts on is a report, not an audit. verify runs the suite on every patch and red goes back to the fixer - that patch only, never the whole batch. the back edge into the map is the whole trick - accepted findings become rules, so next week starts where this one ended. one human step in all of it: which fixes ship - 20 minutes instead of 3 days. save this and read the full graph engineering course below ↓

Hanako

137,040 görüntüleme • 2 ay önce

0:33

an ex-Jane Street trader told me a paycheck is the most overpriced contract there is we were at a rooftop bar in tel aviv. friend of a friend. i told him i work in tech. $4,200/month after taxes. he showed me his phone. a terminal. green numbers. +$47,000. last 30 days. "Claude and one GitHub repo. that's the whole setup" i asked what he means. "copy-trading. but not how everyone else does it" he explained it fast. like he'd said it a hundred times. "most people copy wallets by win rate. useless. a wallet with 85% wins can still lose money. what matters is disposition ratio" he pulled up a wallet. 91% win rate on crypto. 14% on politics. "copy everything - you bleed. copy only their crypto - top 1% of the platform" i went home. couldn't sleep. opened Claude at 2am. "build a copy engine. filter by disposition ratio above 0.70. copy only dominant category per wallet. kelly sizing" gave it this: almost 1k stars. execution engine. order book logic. position sizing. open source. Claude rebuilt the whole system in one night. first week: +$3,200. second: +$2,900. third: +$2,100. fourth: +$1,648. +$9,848 total. 313 trades. 80% win rate. avg hold 5h. the wallets it copies now aren't the highest win rate. they're the best exits. 0xe41f8...b has $43.4K profit. tech only. disposition 0.82. 0xa22c3...d has $40.2K. crypto only. disposition 0.79. 0xf88d2...c has $22.4K. macro only. disposition 0.74. category P&L: > crypto +$974 > weather +$749 > politics +$409 > macro +$353 > sports -$50. killed it capital velocity: 51x. kelly f+ 0.073. drawdown -2.9%. copytrade: saw him two weeks later. showed him the terminal. he scrolled through it. "you built this in one night?" Claude did. "my old desk had 14 people on this" mine costs less than a Netflix subscription. he nodded. "quit your job" i'm thinking about it.

Hanako

311,683 görüntüleme • 6 ay önce

1:40:44

Google Brain founder, Andrew Ng: "Prompting will be dead in 6 months. Loops and graphs are replacing it." in 100 minutes at Stanford he shows how to build agents that finish the work and sharpen themselves LLMs → Agents → Loops → Graphs the first 15 minutes go past where most $500 AI courses stop most people are still learning prompts while the leverage moved two layers up same model, same tokens, and the only thing that changes is the shape you run it in watch the lecture today, then save the full graph engineering guide below

Hanako

122,229 görüntüleme • 2 ay önce

$a Goldman Sachs engineer sat next to me on a delayed flight from JFK four hours on the tarmac. nowhere to go. we started talking. i told him i trade prediction markets with Claude. he almost choked on his ginger ale. "you do WHAT with an AI?" i showed him my terminal. he went quiet. scrolled through it for two full minutes. "this is close to what we run internally. how did you get the scoring model?" i told him. one GitHub repo with 2,600 stars. Claude connected the dots. official Polymarket framework. LLM-powered trading agents. open source. free. i gave Claude the repo and said: build me a grid that scans 700+ markets per hour and enters when the edge exceeds 9 cents. he didn't ask twice. the Goldman guy kept scrolling. "you have a disposition meter. a kelly gauge. a correlation matrix. we spent $4M building this infrastructure" i told him my total cost. Claude $20/mo + VPS $5/mo he laughed. not the funny kind. that was three weeks ago. today the terminal shows: +$16,976. 439 trades. 73% win rate. avg hold 7h. > Falcon Heavy - mkt 68c. model 95c. +$716. > MSFT antitrust - mkt 59c. model 77c. +$670. > Blizzard NY - mkt 74c. model 85c. +$772. > BOJ rate hike - mkt 52c. model 60c. +$422. four bots split everything: - pulse_alpha +$299. - arb_hunter +$558. - trend_rider +$337. - cal_engine +$719. kelly fraction 0.118. correlation matrix tracking 6 asset pairs. when one breaks - something is wrong. the grid trades it. copytrade here: he DM'd me yesterday. "i showed your screenshot to my desk. they want to know who built your infra" i said Claude. he said "seriously" i said seriously.$