luke (@wavegod) header
luke (@wavegod) profile picture

experience building @moonpay | 1% daily in AI, tokenization, & consumer

New York
Joined October 2019
Photo by luke (@wavegod)Photo by luke (@wavegod)Photo by luke (@wavegod)Photo by luke (@wavegod)Photo by luke (@wavegod)Photo by luke (@wavegod)Photo by luke (@wavegod)Photo by luke (@wavegod)Photo by luke (@wavegod)Photo by luke (@wavegod)
Increasingly convinced that the future is human as a company You are the entity and agents run underneath you in sub-orgs. One cluster owns your social media, where an agent sources content, an agent runs each account, an agent handles ads - sourcing leads, negotiating prices, etc. Another cluster owns your job. Revenue, separate from the creator side. A third runs your life. Calendar, taxes, insurance, email, connections. Every workflow has a number on it. Revenue in, cost out. Just like a company balance sheet, except the company is you. You'll one day meet or hear of a 20 year old whose org chart has 25 agents across 5 workflows and a measurable output. The question is not whether this happens, it's when and which humans get there first.
3
76
Allowing users and Enterprises the ability to pipe their own models into your agent harness is a no brainer
2
86
Been thinking a lot about this lately. Building an everything app from a functionality standpoint is becoming increasingly easy. Anyone can integrate rails for memecoins, predictions, perps, and whatever else comes next. Building an everything app that’s actually compelling and retains users is a different challenge entirely. The real moat is the network, not the integration of financial products themselves. You need to attract users, give them meaningful ways to interact with one another (which also enables mechanics like rewards/earnings from callouts, etc.) to build an experience that keeps them coming back. We have spent the better part of two decades learning how difficult this is. From a finance first perspective, there are a couple of crypto native social trading platforms that are very well positioned. From a network-first perspective, it’s X. They are approaching the same opportunity from opposite directions. Interestingly, the walled garden dynamics of social networks still seem as relevant as ever. Open social protocols like Lens and Farcaster would probably be the ideal architecture from a user perspective, allowing people to move between apps without abandoning their personal distribution. But that portability is precisely what makes an open network less attractive as a proprietary moat.
everybody is trying to become the everything app robinhood started with stocks and now is integrating prediction markets & crypto polymarket/kalshi started with prediction markets & now integrating perps & stocks coinbase/kraken started with crypto spot trading & now integrating perps, prediction markets, & memecoins/onchain pump started as a memecoin launchpad & now integrating crosschain trading, perps, prediction markets it seems like whoever has • stickiest customer retention • best rev-sharing systems for new users & whales • cross margining across the stack will win this game dont really have strong thesis on how best to differentiate from others here, also asked this question at the end of coinbase's last earnings call
4
104
WoW Forever is everything I wanted
3
63
I barely even type anymore. My default experience for putting words on a screen is just voice dictating through Granola
2
50
We didn’t come this far just to come this far
BREAKING: $BTC IS NOW NEGATIVE OVER THE PAST 5 YEARS $ETH IS DOWN 61%, $BNB -10%, $XRP -20%, $SOL -70%, $DOGE -77% AND $ADA -92%
3
85
Every AI app has the same onboarding now "What should I call you?" "What are you working on?" "Connect your Gmail" How do we make this better
7
7
207
Tokenized equities will be one of the biggest stories of the next 12 months
BREAKING: The total circulating supply of onchain tokenized equities surged by +$932 million last week, the biggest weekly jump on record. This puts total supply above $4.7 billion, also a fresh record high. Amid the US Midterm elections and the resulting volatility, tokenized equities saw historic volume. In times of volatility, investors are increasingly favoring tokenized securities which trade 24/7. The largest onchain trading platform on Solana, Jupiter, reported +60% MoM growth in the number of tokenized equity traders, with volume up +34%. Furthermore, 62% of that volume now reportedly takes place outside of market hours. Volatility is accelerating the adoption of tokenization.
1
4
96
Tokenized stocks are going to be one of the biggest stories of the next 12 months Onchain stock value went from $639M a year ago to $3.44B today. Plus 4.44M holders. Trading is the obvious way it shows up: - Robinhood stock tokens did $15B in dex volume in their first 100 days - ~1/3 of stock token volume on solana is paired against memecoins - stocks are becoming the quote asset for the attention economy But the number i'm watching is collateral: - only ~2-3% of tokenized stock supply was being borrowed against as of September - Aave's Base equities market went from $0 to $9M+ in stock collateral in 3 weeks, and doubled in 5 days once rewards turned on - Kamino's xstocks lending market is up ~12x since Aug 2025 Meanwhile, Morgan Stanley has tokenized stocks scheduled for Q1 2027. Onchain the product is live (on a saturday, too). Trading will get the headlines, but collateral will prove out ecosystem richness. This year will be wild.
1
24
Muse from @Meta is up five spots on our blended chart this week! #19 last Wednesday, #16 by Friday, #14 today. Big week for @Muse. Meta now has two apps in our top 15, with Meta AI at #6.
2
1
8
3,783
Grok Bot does a great job with user friendliness, no doubt. But even that will get better by Hermes as they continue abstracting away complexity. Enterprises want high levels of customizability when handling their infra as a whole, though, and I don't think they'll stand for not having a hand in their model routing, among other things, which affects cost, performance, and even compliance/regulatory It's an interesting decision tree to think through. In one version of it, Bot adapts to this and begins to look more like a harness with fantastic model routing + customizability under the hood.
Replying to @austin_hurwitz @bot
Might be a hot take but I think that Hermes will actually win enterprise long term unless the others allow for highly granular customizability + cost control.
3
174
Strictly in terms of being a personal agent for keeping track of life, Muse is on top
Muse from @Meta is up five spots on our blended chart this week! #19 last Wednesday, #16 by Friday, #14 today. Big week for @Muse. Meta now has two apps in our top 15, with Meta AI at #6.
9
427
Steal my prompt and make a loop in any art style you want
4
165
Idk who needs to hear this but you cannot just /rea an entire company’s growth playbook, regulatory moat, ops strategy, etc. effectively with one prompt
3
108
Saved you the work Share this post with your AI, swap in your own scene, and it'll build you a cozy loop like this one
Made a tiny PS2-era cozy café loop entirely in code No 3D models, no textures, no AI video. Every prop and pixel is procedural. Opus 5.5, GPT 6.1 Sol, you name it can do this today Here's how to make your own, plus all resources you can save for your agent 👇 1/ Any frontier model. Opus 5.5, GPT 6.1 Sol, you name it. Describe the room, the props and the vibe, and tell it everything has to be built in code. 2/ three.js for the scene. Simple shapes for props, textures generated in code. Personally I asked for a low-res render to get the PS2 look. 3/ Puppeteer to capture it. It runs the scene in a headless browser and grabs every frame so the motion stays smooth 4/ ffmpeg to stitch the frames into a seamless loop 5/ Plain English to iterate. "Warmer light", "more steam off the cup", "fix his arm", etc. The video you see below is a second pass, so it’s very good at giving polished results.
2
75
Everything is a plugin
Your Harness. You Define It. Meet Creator Mode in DeepSeek Harness. Turn ideas into plugins that you can use, from a design workspace to a sandbox game or a 3D LLM visualizer. What will you create?
5
134
DeepSeek's Super-agent paper is one of the best writeups on agent architecture we have seen so far, but if you run a team of Grok Bots, you already have most of it. Always-on cloud, memory, background workers, triggered routines, etc. What's missing still is the discipline around process. In the Deepseek model, this is backed by code whereas in Grok Bot, it is backed by prompts and files that are read at the beginning of each turn. Three initial ideas that may be worth stealing: 1.) Have a single front door; make one bot your router. It talks to you, hands work to your other bots, and never does the work itself. Today, you're likely the router, hopping between chats. That is the entire bottleneck the paper is written about. 2.) Efficiency is not currently a priority by default. We should always think about code before tokens, because every routine wake is a full model run. Put a cheap script check in front ("did anything actually change?") so the bot only wakes when there's something worth doing. This can actually be solved for free by my Scriptoor bot, which I will link in the thread below. 3.) Watch before you trust. Deepseek calls this 'observe' mode. Work you ask for runs right away, but anything a bot decides to do on its own just gets logged as "here's what I would've done" until its calls look right, at which point you let it act. This one is baked into Super Agent through code, but is one that I feel could heavily slow me down, to be fair. TLDR: I am taking notes from this article to improve my existing architecture within Grok Bot so I don't have to babysit a new platform. These are three takeaways I had. There will definitely be MANY more papers just like this one in coming months.
1
2
143
In all honesty, I want to pipe whatever model I want + Hermes directly through Grok Bot as a harness and have that be my stack Would pay monthly just to use the Grok Bot harness, no models underlying tbh
1
5
323