Claude for Hedge Fund Thesis Engine
What this workshop covers
This is a 1-hour hands-on session showing hedge fund PMs, analysts, and research teams how to produce a conviction read and a structured bear case, using Claude AI agents inside Claude Cowork.
We walk you through a production-ready multi-agent workflow that takes a name and a long thesis and produces a tested conviction read with the bear case built against it, so by the end of the session you can pressure-test a thesis without assembling the counter-case by hand.
Limited slots available. First come, first served.
Section 1: Running the Full Thesis Workflow How to have Claude run the entire engine in one sequence. An Orchestrator restates the thesis, the variant perception, and the key drivers. A Long Builder constructs the strongest long case and the price-target logic. A Short Adversary on a fresh turn attacks each pillar of the long and builds the bear case without deferring to it. A Catalyst Tracker maps the events that move the name and their dates. The run ends at the Source Watch Dog, which checks every number to a filing or transcript, confirms catalysts are dated, sends any untied number back to the Long Builder, and has the Judge issue a conviction read, a sizing view, and what would change the mind.
Section 2: Turning the Workflow into a Reusable Skill How to capture the run order as a Claude skill so it executes the same way for every name. We build the skill that encodes each seat, the source-tie discipline, and the catalyst mapping, so producing a conviction read and a bear case becomes a single instruction rather than a sequence you rebuild per ticker.
Section 3: Root Files and Workspace Configuration How the files driving the workflow actually work: the workspace identity and house rules, the skill file for the thesis engine, the editable criteria that define what counts as a sourced figure and how conviction and sizing are scored, and the agent files behind the Long Builder, Short Adversary, Judge, and Source Watch Dog. Configure these once and every name you run gets argued, countered, and source-checked against the same standard.
Section 4: The Scheduling Prompt How to run the Catalyst Tracker live. Configure Claude to monitor filings and transcripts for the names you hold and re-run the bear case when something material drops, so a position's counter-case stays current and you are alerted when the thesis needs revisiting.
What you leave with
All root files, prompts, workflows, and input templates ready to run on your own company tomorrow. 30-minute Q&A with Ayub.
Who this is for
Hedge fund PMs, analysts, and research teams who want to compress the time between forming a thesis and having a tested conviction read with a structured bear case and dated catalysts on the table.