Request an Invitation

July 22, 2026

Dallas AI Club 2.0: Agentic or Degenerate?

By Jeremy Gregg

A watercolor illustration at dusk: a wooden gate set in a low stone wall opens onto a landscape of rolling hills threaded by several winding trails that branch and rejoin as they climb — some going up over a nearby hill, others crossing a stream on stepping stones or curving away toward a hazy city skyline on the distant right horizon.

The night at a glance

  • Returning after a two-week pause to see whether the group still serves a purpose
  • Eleventh meeting. A larger room again — five of us
  • A one-time gathering to explore what we are calling “Dallas AI Club 2.0: Dallas Agentic Investing Club”
  • Watched a tutorial on building a Claude-powered trading agent
  • The real conversation was less about which agent to use and more about what kind of intelligence we actually want inside our investing decisions — speed, or wisdom, or some combination
  • Sharing the research below so members who missed the night have it, and so we have our own record of what we are working through

After the tenth meeting, we hit the pause button — took a couple of weeks off to sit with the question of whether this group still serves a purpose. This eleventh meeting was a one-time gathering to explore what we are calling “Dallas AI Club 2.0: Dallas Agentic Investing Club” — a possible new direction for the room.

Something has been shifting at the Dallas AI Club for a few weeks now, and this Wednesday it settled. The topic of the room is increasingly about how to use AI agents to support investing decisions — not because that is the only interesting application of these tools, but because it is the one where several of us have skin in the game and where the tools have quietly become good enough to actually do the work. Someone in the room jokingly renamed us the Dallas Degenerate Investing Club. Someone else offered Dallas Agentic Investing Club as the more honest version. Both names stuck, and both describe true things about the group. This blog is going to start carrying more of what we are learning as we test these tools, both for our own record and for anyone else on a similar path.

The evening’s centerpiece was a tutorial on building a Claude-powered trading agent, linked at the bottom of this post.

An earlier tutorial from the same channel is worth watching alongside it — it covers connecting Claude to a brokerage account in the first place.

The tutorial is useful. It is also only one of several approaches worth understanding before wiring an AI to actual money. Before the meeting I spent some time with Claude thinking through the actual landscape of what an amateur with somewhere between $100 and $10,000 could reasonably do, and five approaches surfaced that are worth naming.

  1. The boring baseline that wins. Automated recurring buys of BTC, ETH, or index ETFs — built into every exchange and broker, requiring zero code. The AI’s role here is oversight, not trading: monthly portfolio review, drift alerts, rebalancing flags. Unsexy, but it is the benchmark everything else has to beat, and most bots do not.

  2. Congressional copy-trading. An ETF called NANC mirrors Democratic congressional trades and is up about 88.5% since its February 2023 launch versus the S&P’s 74.7% over the same window — a real edge with essentially zero effort. The Republican-trades version has underperformed the market. DIY agent versions of this exist (roughly what the tutorial video builds), but congressional disclosures arrive weeks late by law, so any agent version is always copying old trades.

  3. Grid or range trading on crypto. The “buy at $0.90, sell at $1.10, repeat forever” idea. A real strategy category with real bots behind it. The honest version: profitable in sideways markets, bleeds when the range breaks, no long-term evidence that it beats simply holding the asset. Trading fees eat small swings faster than most people realize; a round trip on Coinbase’s entry tier runs about 1%. Viable as a small experiment, not a core strategy.

  4. Official agent rails. This is the actually new thing. As of this summer, agentic trading stopped being a hack. Coinbase launched Coinbase for Agents — AI agents can trade an isolated portfolio of your actual account with hard caps you set. Alpaca (the brokerage the tutorial video uses) now has an official MCP server, meaning stocks, ETFs, crypto, and options can be traded in plain English from Claude, with a paper-money mode for safe testing.

  5. The AI analyst committee. The most-starred project in this space on GitHub — over 50,000 stars — is an open-source AI hedge fund that runs 19 persona agents (a Warren Buffett agent, a Charlie Munger agent, and so on) that independently analyze a stock, argue with each other, and then a portfolio-manager agent decides. Run it weekly, trade rarely, read exactly how each analyst reasoned. It is not necessarily the most profitable approach, but it may be the best educational artifact in the space — a way to actually see how different investing philosophies produce different conclusions from the same data.

Reading these five approaches side by side is what produced the conversation the meeting actually kept returning to. The question underneath all of this is not how fast can the agent trade? The question is what kind of intelligence do we want it applying? Approach #3 is fundamentally a speed play — the bot is doing something a human could not do at the required velocity. Approaches #1 and #5 are wisdom plays — the bot is holding a longer, more considered thought than the human could hold consistently. Approaches #2 and #4 are somewhere in between.

The interesting shift for several of us has been noticing that the wisdom side is where the actual leverage seems to live. A committee of 19 well-prompted persona agents thinking about a company for an hour is doing something no individual human investor is going to do on their own. That is genuinely new. Speed-trading crypto is not new, and the historical record on it is not encouraging. If DAIC ends up as an investing club, it will probably be one that leans wisdom — using AI to integrate more research and more perspectives than any of us could hold in our own head at once, rather than using it to shave milliseconds off a trade.

For now, we are sharing what we are learning as we test these tools. Nothing in this blog is investment advice. Some of it is the kind of thing you learn only by losing a little money on the way through, and some of the members in the room have already started doing that in small amounts to find out. If you are on a similar path, or if you have ideas of your own, we would love to hear them.

Stay tuned for the next meeting of the degenerates aka the organized idiots.

If you are not yet in the room and want to be, you can request an invitation.

— Jeremy

Sources and further reading

← Back to all posts