Robinhood Chain's $80B DEX Quarter, Grok Bot Puts AI Agents to Work

Vlad said Robinhood Chain launched its mainnet on July 1 and ranked among the top five chains by volume and revenue in its first quarter, with more than $100 billion in DEX volume and over $50 million in chain fees. DeFi perpetuals volume had exceeded $1 billion a day in recent weeks. In Q3, the chain generated $46 million, more than 38% of Robinhood’s $120 million in crypto revenue, which beat the $114 million expectation. Robinhood overall reported 51 cents in EPS versus 56 cents estimated and $1.42 billion in net revenue versus $1.44 billion estimated. A speaker said TVL had risen from zero to $1 billion in three months, citing a post, and said Robinhood was adding staff and engaging projects to support builders.

The host sees the chain as a possible new Robinhood business line capable of generating at least $100 million annually; another speaker predicted it could become the company’s fastest-growing division within a year. Those are forecasts, not reported results. Native tokens rose 30% after the earnings call. The host described the earnings report as a bullish signal for the chain, though did not give a DEX-volume figure. The discussion’s specific volume claim came from Vlad, who reported more than $100 billion in the chain’s first quarter.

Grokbot’s appeal, according to the speakers, is that people can start without special hardware or AI expertise and give agents tasks in natural language. The host expects plug-and-play skills to develop into an app-store-like marketplace, with X creator rewards potentially extending to people whose skills users adopt, not just people who publish posts. Suggested applications included agents that update CRM records after sales calls, prepare proposals, and route unusual deal terms for approval. Other examples covered research, fact-checking, scheduling, audience replies, and managing direct messages, with humans still overseeing or publishing the work.

Claims about agent performance were treated cautiously. One creator said five Grok bots generated one million impressions, including a top post with 154,000 views, but the host said the figure was unverified and could be exaggerated or fabricated. The proposed content workflows ranged from a CMO bot coordinating four specialist bots to a 15-bot operation; Jeremy said one agent session may be able to orchestrate sub-agents, making some multi-bot setups unnecessarily complex. Speakers also described agents summarizing media, monitoring wallets and researching rug risks, doing bookkeeping, and preparing tax records. One speaker said a $20 subscription could replace bookkeeping that had cost thousands, but advised using a tax professional for final filing when significant sums are involved.

The speakers emphasized keeping consequential actions under human control. Drew recommended giving bots and APIs read access first, since read-write access allows them to act on a user’s behalf. One speaker described using read-only Plaid access and personally approving crypto transactions from wallets kept separate from the agent. They recommended having agents present options and wait for approval before transactions or posts, and starting with one task before expanding. A story that a Grok bot had turned $3,000 into $30,000 was explicitly unverified, and the speaker was unsure the technology was ready for that use. Agents and heads-up displays also raised fairness concerns in online poker: on-chain play may not stop cheating if an agent can sign for a player, and the host endorsed forfeiting a match and wager after three failed bot checks.

Reliability and cost were recurring constraints. Jeremy recommended starting agents with manually maintained spreadsheets, then asking them to organize the data or make graphs and dashboards; he also suggested using a second agent to verify sources and maintain a dependable record. A speaker described Prism, a GitHub skill that launches five task-specific agents, including one that challenges assumptions, and said adversarial review made results more trustworthy. One speaker said routing repeatable work to cheaper models and reserving a stronger model for checks at the start and end had cut token use fivefold. That speaker paid $200 a month each for OpenAI and Claude Pro and often hit both plans’ limits. Drew said Grokbot can monitor credit use and shift work at 80%, while warning that automated tasks can consume credits quickly.