After building the research swarm on May 15, I let it run. For the next ten days, I got a morning brief every weekday at 6 AM before the market opened. Here's what I actually learned from the experience — not the picks, but the process.
The Briefings Got Better Over Time
This surprised me more than anything. The first few briefings were good but generic — the kind of market summary you'd get from any financial news aggregator. By the end of the second week, StockPicker was referencing prior conversations, cross-referencing picks it had made earlier, and flagging when a thesis from the previous week had played out (or hadn't).
Agent memory is real and it compounds. An agent that has seen 10 days of your questions and reactions knows your risk tolerance, your sectors of interest, and what level of detail you want. You don't have to tell it these things — it infers them from your pattern of follow-up questions.
The Picks That Got My Attention
The research swarm flagged several interesting setups during this period, including positions in XOM, LMT, JPM, BG, and HAL — all based on specific macro catalysts identified overnight (energy policy shifts, defense budget developments, commodity supply dynamics). I'm not a financial advisor and none of this is investment advice, but it made me a better-informed observer of what was moving markets.
The TechResearcher agent was particularly useful. It flagged a SharePoint zero-day vulnerability before it was widely covered, which mattered for portfolio exposure to enterprise software. Seeing how technical security news ripples into stock price dynamics was a new kind of awareness I hadn't had before.
What a Research Swarm Can't Do
It can't make decisions for you. The briefings are information synthesis, not wisdom. StockPicker explains its reasoning — why a given macro signal points toward a particular sector, why an earnings beat in one company matters for its suppliers — but the actual judgment call is still yours.
I found this reassuring rather than disappointing. I wasn't looking to outsource my thinking; I was looking to start my day with better information. On that front, the swarm delivered consistently.
The Deeper Pattern
Running the research swarm for two weeks taught me something about how to use AI effectively: set it up well once, then let it run and observe. The temptation is to constantly fiddle with the agents, adjust their instructions, tweak the schedule. Resist that. Give the system time to develop a rhythm and accumulate context before you change anything.
When I did make changes — asking TechResearcher to focus more on AI regulatory news in week two — I saw the benefit of that directed focus almost immediately. Small targeted adjustments work better than frequent overhauls.