Creative Agent Use Cases - Anthropic job posts, public CEO brains and competitor checks
Plus: Idea Files for tracking management comp changes and local project resistance
Second Brains
Instead of reading books, the CEO of YCombinator asks his agents to rewrite them as insights mapped to his life (@GarryTan). This is possible because he’s developed a “second brain” - a personal knowledge base making his agents way more useful. Tan shares his framework for maintaining his second brain in the opensource project, GBrain.
What will happen with all these second brains?
Some people want to rent them out. Sequoia-backed Delphi.ai allows influencers to host second brain chatbots based on their essays, interviews and other documents they feed it. Some public company executives like Brian Halligan (Hubspot co-founder/former CEO), Gokul Rajaram (Pinterest and Coinbase board member), Spencer Skates (Amplitude CEO) even have second brains you can talk to.
A couple funny responses from public company executives:
Alfred Lin, Airbnb board member, on their recent decision to commingle hotels with home listings: “The moment you commingle [hotels] into a single search, you force yourself to build one ranking system that has to serve both. And one of them always loses. You either optimize for homes and hotels feel generic, or you optimize for the transaction and homes feel like a commodity. You end up with a muddled middle that serves neither well.”
Mike Ovitz, former Disney CEO, on Disney+ as the “digital centerpiece”: The Fortnite piece is the tell. If Disney's truly centralizing everything on Disney+, why are they planting the flag inside Epic's universe? That's not strategy; that's optionality masquerading as strategy. It signals uncertainty about which direction actually wins.
We’ll be able to benchmark which second brains bring incremental insight to the market. Delphi has bots for four Iran experts - policy analysts and professors - and they all think Polymarket’s current odds on a permanent ceasefire deal are way too high. From political analyst Hooman Majd:
Delphi’s public bots are trained on mostly public info and are mainly for amusement. But the value to building personal knowledge base wikis for your agents is very real. These second brains reflect years of notes, pattern recognition and work products not currently in any model’s training data. Eventually they’ll start interacting with the world in new and exciting ways.
Creative Agent Use Cases
Yet Another Value Blog describes his Claude workflows to run local competitor, discounting and review checks for Sweetgreen (SG) and Cable One (CABO) (Substack)
How to create a synthetic panel of varied investor personas to model how investors will react to any catalyst (LLMQuant)
By deploying a local large language model called Llama 3.1 8B and endowing it with 216 distinct investor personas drawn from the FINRA Foundation’s National Financial Capability Study, the authors generated 1.188 billion synthetic buy, hold, and sell responses to 5.5 million S&P 500 news headlines spanning 2010 through April 2025.
The result is something the finance literature has lacked for decades: a high-frequency, demographically representative measure of how investors actually disagree about specific news, day by day, headline by headline.
YC partner on how and why to switch from Claude Code to Codex (@OMooreTweets)
How Microsoft’s Director of Data Science builds data agents to explain why KPIs are moving (Medium)
To make this work in practice, we made a handful of concrete design choices — each with trade-offs — across architecture, platform, data access, security, and behavioral guardrails
Former Maverick PM discusses his system that identifies three drivers of each company in his coverage and performs 24/7 monitoring on anything related to those drivers from news, sellside research, transcripts, etc (@FundamentalEdge)
University of Zurich paper used LLMs to process 128,860 10K filings to track how firms prepare, insure and financially buffer themselves against physical climate hazards (Climate and Tech)
LLM generated dataset of Canadian miners, comparing production guidance against drill assays, grade-by-thickness economics and regulatory guidelines (Quintarthai)
Jaguar Analytics spends $2,000 per month on Claude and Gemeni to identify hidden angles outside of boilerplate commentary from equity research, earnings transcripts and investor day presentations (@JaguarAnalytics)
Thread from the /LegalTech community about a system to identify shifts in narrative framing (Reddit)
A lot of the difficult reconstruction problems are less about “finding contradictions” and more about detecting shifts in narrative framing over time. Especially with omission cases, the absence itself can become evidentiary: someone stops repeating a prior position... language gradually softens or changes without acknowledgment
Idea Files
Idea Files are detailed prompts you can paste into your Codex or Claude Code to replicate and customize a custom dataset. For more, read here.
How to track what stocks will be impacted by the next Anthropic release (Original Idea: Jeremy Leung, Idea File)
To replicate and customize this with your Claude or Codex, see our idea file.
How to track comp award changes incentivizing M&A (Original Idea: Yet Another Value Blog, Idea File).
“Friday afternoon, the WSJ broke a report that GME is looking to buy eBay…In late November they gave all of their executives’ PSUs that vested only if the company underwent a change of control and the stock was ‘at least $7.50’ per share within the next five years. ... I’m not sure I’ve ever seen a single PSU grant that flashes ‘we are for sale’ harder than that grant.”
We turned this concept into a system that identifies past comp changes that preceded M&A transactions and identifies similar plan updates:
To replicate and customize this with your Claude or Codex, see our idea file.
How to track local friction for power, storage and infrastructure buildouts (Original Idea: PV Magazine, Idea File). The system flagged two recent public-company-linked siting-risk signals:
DTE Energy (DTE)’s Poseyville Solar Park was denied a special-use permit in Ingersoll Township on April 21, 2026
SK Innovation-linked (KRX:096770) Key Capture Energy’s KCE NY 34 project in Saugerties was facing organized opposition, nearly 600 petition signatures, and potential Article 78 litigation as of May 5, 2026.
To replicate and customize this with your Claude or Codex, see our idea file.
Flat Circle mentioned on Bloomberg
Thanks to Justina Lee for mentioning Flat Circle in her piece last week: AI Bots Auditioning for Wall Street Trading Are Mostly Losing. AI trading arenas are public experiments where agents research stocks and prediction markets and make trading decisions, and we track all of them here.
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.