Creative Agent Use Cases - Channel check agents
Plus: where agents are consistently wrong
Where agents are consistently wrong
Claude apparently misled Dan Loeb into thinking Iran could not turn off oil production due to “water coning,” a phenomenon that occurs with too much production (@DanielSLoeb1)
This is an example of LLMs being challenged with highly technical, new investor questions. Claude may have repeated a narrative produced by a political think tank. Just like the market, agents can latch on to well written incorrect information.
We are collecting examples of more content that regularly misleads agents, please reply if you’d like to discuss.
Channel check agents
Financials
To test whether Nubank (NU) was facing non-performing loan stress, an investor used Claude Cowork to scrape Glassdoor, LinkedIn, and Reddit (Jimmy’s Journal)
Identify which banks are growing via new loan agents, a signal for lower creditworthiness (@NDS0909)
Homebuilders
Build a spec-pressure and incentive tracker based on homebuilder community pages (Idea File). Note: “idea files” are long prompts to replicate and customize the analysis in your own Claude or Codex
Retail / Restaurants
Great thread from former Schonfeld / Citadel analyst on using agents to analyze new store productivity and white space opportunity (@GregoryBlotnick)
Turnaround tracker: Agent that monitors employee complaints on reddit for Kohl’s and other retail businesses with new management (Idea File)
Energy / Renewables
Compare E&P management cost-reduction claims against actual filed well budgets (AFE Leaks)
Scan renewable project claims, map to public-company parents, then compare them against newly required NERC IBR registrations (Original idea: PV Magazine, Idea File)
Software
Analyze reddit discussions on new AI products from public software companies (Idea File)
Investment workflows
Stock Thoughts discusses how he uses Claude to manage his PA (Substack)
100 Bagger Hunting describes his agent that scans for “kill conditions” on his data center holding (Substack)
One investor’s prompt for a capital allocation scorecard (AI Investing Lab)
Idea for earnings call agent that extracts management claims then reads competitor transcripts for contradictory claims (@byerlys32)
Interesting prompt for analyzing earnings transcripts (@MashraniVivek)
You are a seasoned analyst who reads between the lines on every management call. Tell me what actually happened:
The 3 real decisions made — not just stated
What management conspicuously avoided answering
The quiet commitments they’ll be held to next quarter
Other news and tools
Synthetic panel startup for hedge funds offers MCP (Thesis Lab). Synthetic panels are collections of agents designed to simulate real human responses. Thesis Lab competes with Aaru, Simile.ai, Electric Twin, Qualtrics and others. See earlier coverage of synthetic panels.
Sellside is increasingly using AI, resulting in increased quality, speed and market reactions (SSRN, thanks to Justina Lee)
Merger arb investors increasingly relying on agents (AI Street)
Follow for more creative agent use cases
If you would like to discuss incorporating agents into your research process, reply to this email or reach out via X or LinkedIn.