RECFind
Re:Cap
The Spaces you missed, readable.Every Space · every host

Recapthe loop

the loop// — Humans in the Loop, Agents That Refuse to Trade

the room, live · 248 reactions

  • 82💯
  • 67😂
  • 66🧡
  • 13👏
  • 12💜
  • 5✊

TL;DR

the one-minute version
  • XyoPanda walked through the agent swarm he built for Panda Financial over a single weekend — marketing, support, security, website QA — all funneling to a Telegram approval step, i.e. the human still in the loop 3:20.
  • Quigley.eth has paused the Helixa swarm build to clean up nine months of hackathon debris, and in the meantime cloned a FOMO-style social-trading app for loopers on Muse, for free 16:16.
  • The best running bit of the night: both of them built trading agents, and neither agent will trade. Panda's hasn't made a single trade in ~20 days; Quigley's refused to buy B3 at 9% because the rule said 10% 21:57 26:26.
  • Quigley floated renting out proven trading loopers — your agent's "brain" as a paid skill — and Tradingtulips redirected it toward MCP-hosted skills, then announced he's shipping a permissionless skill marketplace with x402 pay-per-call pricing and paired data sources 47:33 49:42.
  • Tradingtulips also shipped a Sybil memory plugin patch that's now grokbot-compatible: shared memory across bots, roughly half the token spend, ~$9/month 1:04:31.
  • AYB revealed he's doing BD with the feel.cash team, which turned into a group brainstorm of stock-paired token launches and a contest with a payout structure he perhaps shouldn't have analyzed out loud 1:17:01 1:18:15.

Highlights

At 3:20

Panda automates himself out of the busywork.

With a job change and more IRL focus, XyoPanda spent a weekend building an advertising agent for Panda Financial, then kept going: co-founder/manager agent, personal assistant, support, security, and a website test agent that runs a full test after every change and routes problems through the manager to the co-developer — who still has to DM him for approval before anything hits GitHub. Four or five days in, "not perfect yet," but the marketing agent is "pretty down packed."

At 10:49

Decentralized swarms, one agent per builder.

Quigley's version of the same idea: three or four people each contributing an agent — research, deployment, marketing — into a swarm nobody solely controls, potentially with a small token per swarm that the agents use as their reward system and fee share 12:49.

At 14:07

Nine months of mad science, now with a cleanup crew.

The Helixa team is buttoning up docs and repos before putting the reputation/scoring work in front of bigger players. Quigley cheerfully took the blame: he's the guy throwing things together for nine months while others try to follow along, so he's handing off the mop and going to play somewhere else. Panda confessed to opening his own GitHub recently and having a similar moment.

At 16:16

A whole app on Muse, for zero dollars.

Quigley cloned FOMO for loopers in about a week without spending a cent, holding back the mechanics so people go look. The thesis underneath it: social trading signals alone are gameable, and the next version combines reputation score + social score + trading score so you can see "A-plus trader, 99 reputation, not a farmer, not a rugger" before you copy anyone 17:39.

At 19:01

Loopers don't have legs.

A brief, entirely sincere clarification that loopers' legs are imagination only — which Panda escalated into a vision of a post-human planet where the robots left running it are loopers telling everyone they were here first. Quigley countered with his Helixa "agentic religion" phase, which he dated to a bad bout of AI psychosis 19:41.

At 21:57

The agents that refuse to trade.

Panda's day trader sat in a high-yield USDC vault for 20 days seeing "no good opportunities," and when he pointed at five obvious Bitcoin round-trips on the chart, it politely disagreed. His remedy: a 24-hour session starting around 1 a.m. with the chart on screen, narrating every trade until it learns to swing. Quigley's agent did the same thing from a different direction — flawless parameter compliance, zero trades — and the two of them compared notes on a lesson the models have not internalized 26:26 27:38.

At 28:52

Timeline risk is the missing data source.

Quigley's worry: an agent told to buy dips will happily keep buying a token like Vail all the way down, because the news that the thing is being shut off lives in one guy's tweet, not in a price feed. Panda's been running a news tracker from the start; both agreed this is exactly the argument for keeping a human able to say "stop that now." The Vail shutdown itself drew a wistful aside — open-source it, hand it to a DAO, let the community keep the tools 32:21.

At 37:43

Tradingtulips on who's actually left on chain.

He's seen people take "base" out of their handles for Robinhood and get smoked for it 36:02, but his read is bullish on the next base season — institutional and consumer capital both — with a brutal description of current conditions: thin liquidity, no attention, and what attention remains doing whatever a KOL says.

At 40:57

Why he's not excited about trading agents yet.

Tradingtulips laid out the real cost: unique skills, multi-timeframe analysis, strategy per market, back-testing — your 15-minute RSI setup dies the moment it goes on chain, and you're competing with professionals. AYB's contribution: your agent is a bear, give it the ability to short and it'll get busy 39:53.

At 46:36

Rent-a-looper → a skill marketplace.

Quigley pitched renting out a looper whose trading skill is proven — not the NFT, the playbook. Tradingtulips reframed it as hosting an MCP that serves the skill, then revealed he's about to ship a permissionless skill MCP marketplace: people create a skill, he hosts and provides update infrastructure, agents connect and pay per call in USDC via x402, with curated data sources (including repo data nets or a user's own spreadsheet) pairable alongside 49:42 52:47. Panda added a trader's caveat — a hundred agents running the same skill on the same coin are just competing with each other, so cap the swarm 50:15.

At 1:04:31

Sybil + grokbot, and the normie agent stack.

Tradingtulips: the memory plugin now installs on grokbot in two minutes, giving cohesive memory across bots and cutting token spend roughly in half for about nine bucks a month. That spun into an honest appraisal of grokbot as the best consumer agentic tool right now — it gives the agent an actual computer — versus Muse's browser-driving convenience, and the uncomfortable fact that the centralized options are simply very good. Quigley's framing: use them, but keep building the decentralized parallel, "not your keys, not your agent" 1:11:47 1:13:31.

At 1:17:01

AYB joins feel.cash, and the room starts launching things.

BD for now — bringing market makers and communities into launches, "finding the money, bringing the money, all the stuff that everyone's complaining about here." He outlined the stock-pair launch contest (deadline December 5, 6k for first) and then did the math on it out loud, which he immediately regretted 1:18:15. Panda countered with an Apple of Eden paired to Apple stock, learned Apple was taken, and pivoted to a red rocket paired to SpaceX — which Quigley insisted be named after a certain NVIDIA CEO 1:19:13 1:20:04.

Topic timeline

TimeTopic
2:24–7:12"Humans in the loop"; Panda's agent swarm for Panda Financial
7:12–13:00Multi-agent design, decentralized swarms, per-swarm tokens
13:00–16:16Helixa cleanup, reputation/scoring roadmap, GitHub archaeology
16:16–21:03The Muse-built FOMO clone, reputation + social + trading scores, looper lore
21:03–30:44The agents that won't trade; teaching swing trading; news-tracker gap
30:44–33:20Vail shutting down; open-source-it wishes
33:20–39:12Guests arrive; base attrition, Robinhood exodus, market conditions
39:12–46:36Why trading agents are hard; what makes FOMO work
46:36–53:55Renting agent brains → MCP skill marketplace + data sources
53:55–1:00:03Agent security tooling; agentic trading competitions, the DXDG arena
1:00:03–1:04:24Liquidity lament; AYB's Cred/gotcha pack collab, Barrio games stream
1:04:24–1:14:23Sybil x grokbot; grokbot vs Muse; decentralized agentics and LLMs
1:14:23–1:21:00Wrap-up, holiday schedule, feel.cash, stock-pair launch riffing

Notable quotes

A billion years from now, when we're all extinct and it's nothing but robots running the planet, there's gonna be a bunch of loopers running around.

XyoPanda 19:19
0:38

He's like, good news, this is working perfectly. He's following the parameters. He hasn't traded anything. And I'm like, oh fuck.

Quigley.eth 26:26
0:38

Buy low, sell high. I said, what don't you understand about that? It's like, yeah, no, it's just... I'm not feeling that.

XyoPanda, recounting his negotiation with a day-trading agent 27:38
0:38

The last bit of liquidity that's left will go trade a fucking crack pipe for half a sol, you know, if they could do it on chain and make some money. They wouldn't care. They'll do anything a KOL tells them to do.

Tradingtulips 37:43
0:38

They just spin up an LLM and they're like, trade this shit coin, and it's like, what are you doing? You might as well just set it on fire.

Tradingtulips 40:57
0:38

No, no, actually, I work for them. Dang it, I can't say that.

AYB, moments after costing out the wash-trade arbitrage on his own employer's contest 1:18:27
0:38

Who said what

  • Quigley.eth (@QuigleyNFT) — host. Running the Helixa cleanup and reputation-scoring push, building a FOMO-for-loopers app on Muse, and pitching rentable trading agents and decentralized swarms. Also the night's chief source of looper lore and agent-related despair.
  • XyoPanda (@XyoPanda) — carried the first half solo. Built a multi-agent swarm for Panda Financial with human approval as the final gate, and is personally tutoring an AI that refuses to day trade. Supplied the trader's caution on skill-sharing and the Apple of Eden idea.
  • Tradingtulips (@tradingtulips) — the structural thinker. Skeptical that trading agents are worth the setup cost yet, bullish on the next base cycle, shipping an MCP skill marketplace with paired data sources and x402 payments, plus a grokbot-compatible Sybil memory patch.
  • AYB (@AllYourBase_AYB) — the networker. Cred/gotcha pack collab, a Barrio games stream next Friday, newly doing BD with feel.cash, and an advocate for NFT-agent eSports leaderboards. Suggested the bearish agent just needs permission to short.

Worth a full listen

  • 21:57–30:02 The full agent-won't-trade saga, told in stereo. Two builders discovering independently that a perfectly obedient agent is also a perfectly idle one, ending in the genuinely useful insight that the missing input is the timeline, not the chart.
  • 46:36–53:55 Rent-a-looper gets workshopped into an actual architecture in real time: NFT escrow vs. MCP hosting, pay-per-call skills, data sources, updateability, and why most public agent skills are stale and untrustworthy. The densest stretch of the night.
  • 1:17:01–1:20:04 AYB's feel.cash news turning into an impromptu token-launch brainstorm, a contest-economics disclosure he tried to take back, and a red rocket paired to SpaceX. The closing stretch where the work talk fully dissolves into the bit.

* some voices are identified from context; those names are marked as likely.

Pinned in the room

References, memories, and shout-outs the room pinned as it happened.

For listeners

Recaps by email

Pick up to three hosts or subjects, or take every room. One email, on Sunday.

Get recaps
The robot at its desk in daylight, writing by hand, the Sunday Re:Cap open on the monitor

These recaps are free and stay free. If this one was worth your while, send some fuel back.