Jamie Dimon Built a Full Treasury Dashboard With Claude in 20 Minutes. Here Is What He Used.
On May 5, 2026, Jamie Dimon stood on a stage in New York next to Dario Amodei and described what he had done the weekend before.
He had logged into Claude Code on his own. Typed in what he wanted: asset swaps, Treasury bid-ask spreads, investment grade analysis. In twenty minutes, Claude had built him a full dashboard with backup data, research, and analysis accurate enough to use in production.
The CEO of JPMorgan. Twenty minutes. On his own.
That is the context for what Anthropic launched that same day: ten AI agent templates for financial services, built on Claude Opus 4.7, connected to the data infrastructure that Wall Street firms pay six figures a year to access. FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, LSEG, Moody's, and more.
Most of these agents are designed for institutional teams. Banks, hedge funds, asset managers deploying Claude at scale across their entire workflow.
But three of them are immediately useful for any individual investor who wants to do serious research without a Bloomberg terminal or a financial advisor.
Here is what they are, how to set them up, and what they actually produce.

What Anthropic Actually Launched

The ten agents are reference architectures, not basic prompts. Each one packages three things: skills (the domain-specific instructions and knowledge for the task), connectors (governed access to financial data sources), and subagents (specialist Claude models called for specific sub-tasks like comparables selection or methodology checks).
The agents deploy as plugins in Claude Cowork, as skill bundles in Claude Code, or as cookbooks for Claude Managed Agents running headlessly on Anthropic's infrastructure.
The ten agents split into two categories:
Research agents (the ones useful for investors):

• Market Researcher
• Model Builder
• Earnings Reviewer
• Pitch Builder
• Meeting Preparer
  Operations agents (designed for finance professionals):
• Valuation Reviewer
• GL Reconciler
• Month-End Closer
• Statement Auditor
• KYC Screener
  These updates pair best with Claude Opus 4.7, which leads the Vals AI Finance Agent benchmark at 64.37%, the best of any frontier model on financial tasks.

Setting Them Up

Start with the Market Researcher. Not the Model Builder.
The Market Researcher requires the least configuration, produces useful output on the first run, and immediately shows whether your setup is working. If the Market Researcher returns a clean portfolio briefing with real analyst data and recent news, the installation is confirmed and the other two agents will work too. If it returns nothing or errors, troubleshoot here before adding the others.
Install the Model Builder second, after you have run at least one successful Market Researcher session. The Excel output adds a step that requires either the Microsoft 365 Claude add-in or a manual download, so it is worth confirming the core data access is working first.
Add the Earnings Reviewer last. It is the most targeted of the three and works best once you have already used the Market Researcher to identify which positions need thesis checking.
There are two ways to install the agents. The easiest is through the Financial Services marketplace inside Cowork. The more configurable path is through Claude Code.
Via Cowork (easiest):

Open Claude Cowork on desktop In the left sidebar, click Plugins Search for Financial Services marketplace Click Add Marketplace Then install the specific agents you want from the marketplace
Via Claude Code (command line):
Add the marketplace first: claude plugin marketplace add anthropics/financial-servicesInstall the core financial analysis skills (required first): claude plugin install financial-analysis@claude-for-financial-servicesThen install the individual agents you want: claude plugin install market-researcher@claude-for-financial-services claude plugin install model-builder@claude-for-financial-services claude plugin install earnings-reviewer@claude-for-financial-services
You need a paid Claude plan to use plugins and Cowork. Here is what each tier costs and what it gets you:

• Pro ($20/month, $17/month annual): includes Cowork, plugins, and the Finance agents. The right starting point for most individual investors.
• Max 5x ($100/month): same features as Pro, five times the usage headroom. Upgrade here if you hit Pro limits running multi-agent research sessions.
• Max 20x ($200/month): twenty times Pro's headroom. For daily heavy use.
• Team ($25/seat/month standard, minimum five seats): for organizations.
  The agents themselves are free. Plugins and Agent Skills do not cost extra on any paid plan. Some of the premium data connectors (FactSet, S&P Capital IQ, PitchBook) require separate subscriptions from those data providers. Most individual investors will get substantial value from the publicly available data alone: SEC filings, news, earnings transcripts, and free financial APIs.
  Once installed, the agents appear in Cowork dispatch, skills fire automatically in context, and slash commands become available in your sessions.
  The slash commands are typed directly in the chat window, not as menu selections. Here is what each one triggers:
• /comps — runs a comparable company analysis for the ticker you specify next
• /dcf — builds a discounted cash flow valuation for a named company
• /earnings — pulls and summarises the most recent earnings call transcript
• /ic-memo — drafts an investment committee memo for a position you describe
  Type /comps NVDA and the agent runs comparables against NVDA's peer group immediately. No additional prompt required. These commands are shortcuts for the most common tasks. For anything more specific than what the command covers, use the natural language prompts in the sections below.

Agent 1: The Market Researcher

Your personal analyst covering every stock you own, pulling every relevant data point automatically.
The Market Researcher tracks sector and issuer developments, synthesizes news, filings, and broker research, and flags items for review. For an individual investor that translates to: every analyst rating change, every SEC filing, every major news story across your entire portfolio, organized and surfaced without you going looking for it.
What it monitors:

• News flow, filtered for what actually affects your positions
• Analyst rating upgrades and downgrades with reasoning
• SEC filings parsed into plain English (10-K, 10-Q, 8-K)
• Sector-wide developments that affect your holdings
• Competitive moves by companies adjacent to what you own
• Risk signals: regulatory changes, lawsuits, management changes
  What "publicly available data" actually means without a premium connector subscription:
  Without a FactSet or S&P Capital IQ subscription, the Market Researcher pulls from SEC EDGAR (all public filings), free financial data APIs (price history, basic fundamentals), web-accessible news, and earnings transcripts from company investor relations pages. This covers most of what an individual investor needs for thesis research.
  What it cannot access without a premium subscription: real-time analyst consensus estimates, proprietary broker research reports, and institutional-grade alternative data. If you ask for analyst consensus and Claude cannot access the paid source, it will tell you and work from what is available rather than fabricating numbers.
  Portfolio briefing prompt:

  I own shares in AAPL, NVDA, MSFT, GOOGL, and AMZN.Give me a full briefing on all five stocks. What happened this week? Any analyst rating changes? Any news I should know about? Flag anything that changes the investment thesis for any of these positions.
  Single stock deep dive:
  Give me a complete research report on Tesla (TSLA). Cover: recent earnings, analyst consensus, key risks, competitive position, upcoming catalysts, and whether the current valuation makes sense based on growth.
  Sector scan:
  I am interested in the AI chip sector. Give me an overview of the key players, market share, revenue growth, and which companies are best positioned for the next two to three years. Include NVDA, AMD, INTC, AVGO, and any others worth watching.
  Weekly watchlist ranking:
  Here is my watchlist: PLTR, SNOW, CRWD, NET, DDOG.Rank them by investment attractiveness right now. For each: current price context, recent developments, analyst sentiment, and the one thing to watch most closely.
  The most useful setup: give it your full portfolio once and ask for a weekly briefing every Monday morning. The same prompt, run every week, keeps you current without manually checking each position.

Agent 2: The Model Builder

Full financial models that investment banks charge clients thousands of dollars to produce.
Give the Model Builder any publicly traded company and it produces a complete model: revenue and expense projections, DCF valuation, comparable analysis (P/E, EV/EBITDA, P/S), sensitivity tables across different growth and margin scenarios, and a clean assumptions page. In Excel, formatted for review and adjustment.
This is the agent that directly replaces the tool individual investors have never been able to access without paying an advisor or subscribing to a professional research service.
A note on the Excel output before running the first model:
The Model Builder produces a downloadable Excel file. How it reaches you depends on your setup.
If you have the Claude Microsoft 365 add-in installed (available at claude.ai/integrations or through your organization's Microsoft 365 admin), the model opens directly in Excel with context carrying automatically between Claude and the spreadsheet. This is the cleanest version of the workflow.
If you do not have the add-in installed, the model is generated and offered as a file download inside the Cowork window. Click download, open it in Excel or Google Sheets, and review the assumptions tab before interpreting anything else.
Either way: review the assumptions page first. The model is populated from public data but the growth rates, margin assumptions, and discount rate are defaults that may not reflect your own view of the company.
Full model build:

Build me a complete financial model for Microsoft (MSFT) in Excel. Include five years of revenue and expense projections, a DCF valuation, a comparable analysis versus peers, and a sensitivity table. Show me what the stock is worth at different growth and margin assumptions.
Two-stock comparison:
I am deciding between CrowdStrike (CRWD) and Palo Alto Networks (PANW). Build me a side-by-side model comparing revenue growth, margins, valuation multiples, and implied upside for both. Which one is cheaper relative to its growth?
Scenario analysis on an existing position:
I own Amazon (AMZN). Build me a model that shows what happens to the stock price if: AWS growth slows to 15%, AWS growth stays at 25%, AWS growth accelerates to 35%. Hold everything else constant.
One honest point on using this: financial models are only as good as their assumptions. The Model Builder produces the structure and populates it from public data. Review the assumptions it makes before drawing conclusions from the output. Adjust the assumptions that do not match your own research. A model that you have stress-tested is a useful tool. A model you have not looked at is just a formatted spreadsheet.

Agent 3: The Earnings Reviewer

Earnings calls are where companies tell you what is actually happening. They are also long, dense with corporate language, and most investors cannot listen to twenty calls a quarter.
The Earnings Reviewer reads the full transcript and delivers what matters: what beat and missed, what changed in guidance, whether the thesis you are holding the stock around still holds, and how management's tone has shifted across multiple quarters.
Earnings recap:

Apple (AAPL) just reported earnings. Review the full transcript. What beat and missed expectations? What changed in guidance? Is there anything I should be concerned about?
Thesis check (the most useful version of this prompt):
I own Nvidia (NVDA) because I believe AI infrastructure spending will continue to grow for at least three more years. Review their latest earnings call and tell me: does management's commentary support or undermine this thesis? Be specific.
Here is the kind of output this produces when it works correctly:
Example Earnings Reviewer output
Thesis assessment: Supports with one material complication.Supporting: Jensen Huang described data centre demand as "insatiable" and said the transition from general-purpose to accelerated computing is happening "much faster than we expected." Gross margins expanded to 78.4%, indicating no pricing pressure despite accelerating supply. Forward guidance came in above consensus on both revenue and gross margin.Complication: Management flagged that sovereign AI spending, meaning governments building their own national AI infrastructure, is becoming a significant portion of demand. If sovereign spending is pulling forward demand that would otherwise occur over three years, the growth runway you are modelling may front-load rather than extend. This is not a thesis breaker but it is worth tracking quarter over quarter.Net: The three-year AI infrastructure growth thesis is supported by this call. The one question worth adding to your tracking list: what percentage of current demand is sovereign versus commercial, and whether that mix is shifting.
That is the difference between reading the earnings call yourself and running it through a thesis-specific prompt. The output is calibrated to your position, not a generic summary.
Multi-quarter tone analysis:
Pull the last four earnings calls for Meta (META). How has management's tone changed over the past year? Are they getting more or less confident about their AI investments? Is ad revenue re-accelerating?
Earnings season batch:
This week, AAPL, AMZN, GOOGL, and META all reported earnings. Give me a summary of each. Which beat expectations? Which missed? Which had the most concerning guidance? Rank them from most bullish to most bearish based on earnings quality.
The earnings reviewer works significantly better when you state your thesis first. "I own this stock because of X. Does this earnings call support or undermine that?" produces targeted, useful analysis. "Review this earnings call" produces a summary you could have read in any financial news outlet.

Chaining All Three

The real value emerges when all three agents run in sequence in a single conversation.
Full due diligence on a new position:

I am considering buying shares of Palantir (PLTR). Run full due diligence:1. Market research: news, analyst ratings, competitive position, key risks 2. Build a full financial model with DCF valuation and sensitivity table 3. Review the most recent earnings call and tell me if the thesis holds 4. Final assessment: buy, hold, or pass, with reasoning
Pre-earnings preparation:
Nvidia reports earnings next week. Help me prepare:1. What are consensus estimates for revenue and EPS? 2. What are the three things to watch most closely? 3. Build a quick model showing what the stock is worth at different revenue growth scenarios 4. What did management guide for this quarter on the last call?
Stock screening:
Find me five stocks I am probably not looking at that meet these criteria: market cap between $5 billion and $50 billion, revenue growth above 25%, improving margins, strong competitive moat. For each one, give me a quick overview of why it is interesting and what the risks are.
Each agent builds on what the previous one produced. Research informs the model assumptions. The model gives context for interpreting what management says in the earnings call. The earnings call tells you whether the research thesis is holding.

What the Other Seven Agents Are For

The remaining agents are built for finance professionals and institutional teams:

• Pitch Builder — creates target lists, comparable analysis, and drafts pitchbooks for investment bankers
• Meeting Preparer — builds client and counterparty briefing documents before calls
• Valuation Reviewer — reviews valuations against comparables and methodology for consistency
• GL Reconciler — reconciles general ledger accounts and performs NAV calculations
• Month-End Closer — manages the month-end close workflow and task sequencing
• Statement Auditor — reviews financial statements for internal consistency and audit readiness
• KYC Screener — handles Know Your Customer document review and compliance preparation
  These are functional and genuinely useful for their intended audience. For an individual investor, the first three are the ones that matter.

The Honest Limits

These agents produce research and analysis. They do not execute anything. They do not predict the future. They do not replace judgment on whether a thesis is right.
What they do replace is the time cost of pulling data, structuring models, and reading earnings transcripts manually. The analysis that comes back requires the same evaluation you would give any research output: check the assumptions, verify the data sources, apply your own judgment to the conclusions.
Jamie Dimon built his dashboard in twenty minutes. He then presumably applied twenty years of context in interpreting what it showed him. The twenty minutes is what Claude gave him back. The interpretation was still his.
These agents give you the same compression. What you do with the output is still the job.
The full repository is at github.com/anthropics/financial-services. Everything is open source.
This article covers AI research tools, not investment advice. Nothing here is a recommendation to buy or sell any security. Always do your own research and consult a qualified professional before making investment decisions.
Follow @damidefi on X for daily Claude AI tools, crypto analysis, and the full journey to 100K. Bookmark this. Share it with one person paying a financial advisor for research Claude can now run in under five minutes.