A prompt does one thing, once. An agent does a job, on repeat — it follows your instructions, reaches into your data, and runs a workflow end to end. This floor shows you what finance agents are, when to reach for one, and gives you build-ready specs for the desk, the deal team, and the back office.
Prompt vs. Agent
A Prompt
One Question, One Answer
You ask, it answers, you're done. Perfect for a single task you can describe in a message.
Lives in a single chat turn
You provide the context each time
No memory of your standards between sessions
You do the copy-paste and the follow-up
Reach for a prompt when: it's a one-off, or the task changes every time.
An Agent
One Job, On Repeat
You configure it once with instructions, data access, and a defined output — then it runs that job the same way every time.
Holds your instructions and house style
Connects to your files, data, and tools
Can run a multi-step workflow autonomously
Produces a consistent, reviewable deliverable
Reach for an agent when: you do the same task weekly, to a fixed standard, on living data.
Both Microsoft and Anthropic converged on the same anatomy. Once you see the three parts, every agent template makes sense.
Instructions / Skills
The how-to. Plain-language rules and domain knowledge — "build the DCF this way, flag these risks, format like this." In Claude these are reusable Skills (markdown files); in Copilot, the agent's instructions.
Connectors / Knowledge
The data. Governed access to your sources — SharePoint, Excel, a CRM, or market data via FactSet, S&P Capital IQ, PitchBook, Moody's. Connected through MCP (Claude) or knowledge sources (Copilot).
Subagents / Workflow
The hands. Steps the agent runs to finish the job — and specialist subagents it calls for sub-tasks like comps selection or a methodology check. A human signs off on every output.
Where You Build Them (2026)
Microsoft Side
Copilot Agents
Agent Builder (in Copilot Chat)No-code, conversational. Build an agent for yourself or a small team from your SharePoint, files, and email. Start here.
Copilot StudioThe advanced tier — richer customization, more connectors, governance, and multi-agent workflows. Copy an Agent Builder agent up into it when you outgrow the basics.
Agent 365The control plane IT uses to inventory, secure, and govern every agent in the org.
Best fit if your firm runs on Microsoft 365 and you want agents living inside Teams, Outlook, and Office.
Anthropic Side
Claude Agents
Agents for Financial ServicesTen ready-to-run finance agent templates — pitch builder, meeting preparer, market researcher, KYC screener, month-end closer, and more — each packaging skills + connectors + subagents.
Claude Cowork & Claude CodeRun the agents as plugins on paid plans, or as Managed Agents on the Claude Platform for programmatic use.
Add-ins + MCP connectorsClaude works natively inside Excel, PowerPoint, and Word, drawing on governed market-data connectors (FactSet, Capital IQ, Morningstar, Moody's, and more).
Best fit for deep financial modeling and research workflows that span Excel, slides, and live market data.
Claude's Finance Agents, Explained
On May 5, 2026, Anthropic shipped ten ready-to-run agent templates built specifically for financial services — the most time-consuming work on the desk and in the back office. These are real, named templates you can put to work in days, not months. Here's what they are, how they're built, and where you'd actually use them.
Research & Client Coverage
Front-Office Agents
Pitch builderCreates target lists, runs comparables, and drafts pitchbooks for client meetings.
Meeting preparerAssembles client and counterparty briefs ahead of calls.
Earnings reviewerReads transcripts and filings, updates models, flags thesis-relevant changes.
Model builderCreates and maintains financial models from filings, data feeds, and analyst inputs.
Market researcherTracks sector and issuer developments; synthesizes news, filings, and broker research for risk review.
Finance & Operations
Middle & Back-Office Agents
Valuation reviewerChecks valuations against comparables, methodology, and the firm's review standards.
GL reconcilerReconciles general-ledger accounts and runs NAV calculations against the books of record.
Month-end closerRuns the close checklist, prepares journal entries, and produces close reports.
Statement auditorReviews financial statements for consistency, completeness, and audit-readiness.
KYC screenerAssembles entity files, reviews source documents, and packages escalations for compliance review.
How Each Agent Is Built — Three Parts
1 · Skills
The instructions and domain knowledge for the task — e.g. the KYC screener ships with a "kyc-rules" skill that encodes how to apply your firm's KYC/AML rules.
2 · Connectors
Governed, real-time access to the data the task runs on — FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, Moody's, and your own warehouses and CRMs.
3 · Subagents
Additional Claude models the main agent calls for specific sub-tasks — like comparables selection or a methodology check — so the work is decomposed cleanly.
Two Ways To Run Them
Alongside You
As a Plugin in Cowork / Claude Code
The template runs next to the analyst, using the software already on the desktop. Hand the Pitch agent a target list and you get back a comps model in Excel, a pitchbook drafted in PowerPoint, and a cover note ready in Outlook — all from one request.
On Its Own
As a Claude Managed Agent
The same template runs autonomously on the Claude Platform — for work that spans a whole book of deals or a nightly schedule. Comes with long-running sessions, per-tool permissions, managed credential vaults, and a full audit log where compliance can inspect every decision.
Where You Actually Use Them In Claude
Claude CoworkThe knowledge-work app for non-developers. Install agent templates as plugins; assign work by text or voice with Dispatch, even while you're away from your desk.
Claude CodeThe same templates run here for programmatic, developer-led workflows.
Claude in ExcelBuilds models from filings and feeds, audits formulas across linked workbooks, runs sensitivity analyses.
Claude in PowerPointDrafts decks that update automatically when the underlying numbers change.
Claude in WordEdits credit memos against your firm's own templates.
Claude in Outlook(Now available via the Claude add-in for Outlook) Triages the inbox, arranges meetings, and drafts responses in your voice.
The Claude PlatformRun agents as Managed Agents for autonomous, scheduled, or book-wide work with a full audit trail.
The human stays in charge. In every mode, you review, iterate on, and approve Claude's work before it reaches a client, gets filed, or is acted on. The agent does the assembly and the first draft — the professional owns the judgment, the relationship, and the sign-off. Access requires a paid Claude plan; templates live at Anthropic's financial-services marketplace, and these capabilities pair best with Claude Opus 4.8, state-of-the-art on finance tasks.
The Agent Gallery
Each card shows where the agent comes from and exactly how to use it. Some are real Anthropic Claude templates you install in Claude Cowork or Claude Code; others are build-your-own blueprints you stand up in Copilot Agent Builder or as a Claude custom agent. A few names (like "KYC Screener") are Claude templates and won't appear in the Microsoft Copilot store — the card tells you which is which.
The One Rule That Doesn't Bend
A Human Signs Off On Every Output
An agent is a tireless analyst, not a decision-maker. Never let an agent place a trade, send a client deliverable, file a report, or move money without a person reviewing and approving it. Build the human checkpoint into the workflow — and keep the agent's data access scoped to exactly what the job needs, nothing more.
Build Your First Agent
Master the GCES method first — the same four-part formula that makes a great prompt makes a great agent instruction. Then practice live in the War Room.