Claude & Perplexity research
Earnings summarization, CIM and filing extraction, and precedent search with strict source-citation discipline. Largest leverage, real danger: verify what comes back.
Forge · AI Toolkit
Wield AI like a trained operator, not a tourist.
The research, automation, and agent workflows reshaping the desk, taught with the judgment to direct them and the guardrails to catch what they get wrong.
Why it matters
The value moved from 'I grind longest' to 'I frame, direct, and decide best.'
The desk is changing under your feet. The question is whether you drive the change or get absorbed by it. The people who learn to frame and direct first are the ones who get promoted first. This is the most direct return on a few hours of your week we can offer.
JPMorgan put AI in front of about 250,000 staff, and roughly 72 percent of one Morgan Stanley intern class report using ChatGPT daily. Early fluency is a promotability signal while it is still rare, and a baseline expectation soon after.
Source: Knowledge / AI Capability Assessment 2026
/Hours back, and you can defend every one.
Scenario automation saves 3 to 5 hours a week, AI around the edges another 2 to 4, and you can stand behind every output you ship.
/Get ahead of the rollout, not chasing it.
Learn the workflows your desk is adopting before they become table stakes.
/Speed that compounds on judgment.
First-draft comps in 50 to 70 percent less time, then spend the reclaimed hours on the calls that move the deal.
/Know exactly where the line is.
MNPI, audit trails, and the governance boundary are taught explicitly. You leave knowing what you can paste and what you cannot.
What's inside
Three workflows and one map that tells you when to trust each tool.
Excel & Python automation
Build scenario toggles and run a model audit. The tool catches most mechanical errors; you stay the one who catches the edge cases it misses.
Agent workflows
n8n and Claude pipelines for comps, precedents, and IC-memo first drafts, each wired with human checkpoints. First drafts and copilots, never the final word.
The capability map
Green (first-draft), yellow (copilot), red (human-required: normalization, synergies, capital structure, deal narrative). That map is the difference between an operator and a tourist.
Proof and credibility
We sell the honest frame, because the honest frame is what makes you safe. The best frontier model scored about 64 percent on a finance-agent benchmark and still underperforms a strong junior on judgment. It hallucinates figures, breaks on circular references, and stays blind to EBITDA normalization. The Toolkit teaches the leverage and the catch in the same breath.
Source: Knowledge / AI Capability Assessment 2026 · Perplexity research, 2026
Straight answers
Before you ask.
Will this make my modeling skills atrophy?
Only if you stop thinking, and we are built to keep that from happening. AI sits on top of fundamentals as a multiplier, never a shortcut. That is the competency-collapse risk we designed against.
My firm bans public AI tools, is this useless?
No. We teach the governance boundary and enterprise-appropriate patterns, not 'paste the CIM into a public chat.' The judgment travels to whatever tooling your firm approves.
Isn't this just prompt tricks?
No. It is the capability-tier model, the documented failure modes, and a verification discipline you can apply to any tool. Prompts change; the judgment does not.
I'm a student, can I access it?
In part. A free preview module shows where AI and finance meet. The full Toolkit is part of the professional subscription, which keeps the student tier free.
