This morning Anthropic shipped ten ready-to-run agent templates for financial services and insurance. The lineup reads like the org chart of a mid-sized investment bank: Pitch builder, Meeting preparer, Earnings reviewer, Model builder, Market researcher, Valuation reviewer, General ledger reconciler, Month-end closer, Statement auditor, KYC screener. Each one ships as a Claude Cowork plugin, a Claude Code plugin, and a Claude Managed Agents cookbook — three different deployment shapes for the same template.
📖 Read the full version with charts and embedded sources on AgentConn →
The HN thread (item 48023533, 101 points) hit the front page within hours. Bloomberg, Fortune, The Register, and The Decoder all ran the launch — unusual symmetric coverage that tells you the AI press considers this an IPO-runway move, not just another product drop. Anthropic also signed Microsoft 365 and Moody's data partnerships the same day. Same morning TradingAgents picked up +2,415 stars (now 69K total) and Dexter added +660. Today's GitHub trending is all finance agents.
This piece is the buyer's guide. We mapped each Anthropic template to the actual finance team that already does the work manually, paired it with the closest open-source counterpart so you can see the hosted-vs-self-host trade-off, and flagged where the Sierra-style $15B vertical-CX play is the correct read versus where it's overpaid hype. If you read our piece on the vertical agent wave from May 3, this is the closed-source enterprise companion — same thesis, opposite buyer.
The Architecture Anthropic Is Selling
Before the table, the structural choice. Each Anthropic template is not a single prompt — it's a packaged reference architecture with three layers:
-
Skills — instructions and domain knowledge encoded as
.mdfiles Claude reads on demand. The KYC screener, for instance, ships with akyc-rulesskill that spells out how to apply a firm's KYC/AML rules to a parsed onboarding record. - Connectors — governed access to data the task runs on. The launch lists FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, Chronograph, LSEG, Daloopa, and the new Moody's partnership.
- Subagents — additional Claude models the main agent calls for specific sub-tasks (comparables selection, methodology checks, document review).
The three deployment shapes matter for procurement:
- Claude Cowork plugin: runs in the analyst's browser alongside Excel/PowerPoint/Outlook. Human-in-the-loop. Best for: pitchbook drafting, earnings prep, model review.
- Claude Code plugin: runs in the analyst's terminal. Same template, more code-aware. Best for: model-building, valuation auditing, scripted reconciliation.
- Claude Managed Agents: runs unattended on Anthropic infra. Scheduled or event-driven. Best for: nightly month-end close, batch KYC review, document audits.
The pricing isn't published per-template, but Anthropic's existing Cowork seat ($60/user/month) plus per-agent run charges plus connector data fees is the public template. Per InvestmentNews, enterprise contracts are quoted in the high-six- to low-seven-figure range for full ten-template deployment with custom skills.
The Job-to-Be-Done Map
Each template, the team that owns the workflow today, and the open-source counterpart you should benchmark against if you're tempted to self-host.
1. Pitch Builder — IB Coverage Teams
Anthropic's job: Produce a comps model in Excel, a pitchbook drafted in PowerPoint, and a cover note in Outlook. Triggered by a deal lead saying "we need a pitch for X by Friday."
Who does it today: Junior analysts, ~12-hour turnaround, with a senior associate sanity-check.
Open-source alternative: None close. The pitchbook genre is too tied to firm-specific templates and Microsoft Office formatting nuance — the moat here is the Office connectors, not the LLM.
Buy or build: Buy. The Office wiring alone is a quarter of engineering you don't want to do.
2. Meeting Preparer — Coverage / Sales
Anthropic's job: Pull the client's recent filings, news, your firm's last interactions, and produce a pre-meeting brief.
Who does it today: Analyst spends 90 minutes on it the morning of the meeting.
Open-source alternative: The DIY pattern is a LlamaIndex RAG pipeline over your CRM + filings + news. Real, but takes a quarter to ship and 18 months to make trust-grade.
Buy or build: Buy if you already have Cowork. Build if you have a senior data engineer with three months free.
3. Earnings Reviewer — Equity Research
Anthropic's job: Analyze earnings transcripts, flag tone changes, surface guidance shifts, draft your firm's update note.
Who does it today: Sell-side junior + senior analyst. Two-day turnaround for a full update note.
Open-source alternative: TauricResearch/TradingAgents — their Bull Researcher and Bear Researcher subagents debate over fundamentals, news, and sentiment, then a Trader synthesizes a position. Different shape (multi-agent debate vs. single-agent draft), but the underlying capability is the same.
Buy or build: Build with TradingAgents if your output is a trade thesis (TradingAgents was designed for this). Buy Anthropic's if your output is a client-facing update note (Anthropic's writes prose; TradingAgents writes positions).
4. Model Builder — IB / PE Modeling
Anthropic's job: Build a DCF, LBO, or merger model from financials. Iterate on assumptions. Hands a clean Excel file back.
Who does it today: Investment banking associate. 6–10 hours per model, 3–4 hours of senior review.
Open-source alternative: None at parity. Excel-fluency is the gap. Generic LLMs can write the formulas, but the structural conventions of a Wall Street model are tribal knowledge, not training data.
Buy or build: Buy. The Excel fidelity is the entire feature.
5. Market Researcher — Coverage / Strategy
Anthropic's job: Stand up a research brief on a sector, theme, or specific company on demand. Pulls from FactSet, S&P, IBISWorld, Third Bridge expert transcripts.
Who does it today: Strategy associate, 4–6 hours.
Open-source alternative: virattt/dexter — autonomous deep-financial-research agent in TypeScript, includes a LangSmith eval suite and LLM-as-judge scoring. Real product, designed for exactly this job, supports OpenAI and Anthropic models.
Buy or build: Build with Dexter if you have engineering, want full control of data sources, and can BYO connector to your firm's research repository. Buy Anthropic's if you don't, and you want the Third Bridge / Guidepoint expert-network connectors out of the box.
6. Valuation Reviewer — IB / PE Quality Control
Anthropic's job: Audit an existing model. Flag formula errors, broken references, inconsistent assumptions, comparable selection issues.
Who does it today: Senior associate / VP. 2–4 hours of senior time per model.
Open-source alternative: None. The pattern of "spot what's wrong with a Wall Street model" is rare-and-tribal training data — and the Excel fluency gap is the same as Model Builder.
Buy or build: Buy. This template is specifically the kind of senior-time-saver that justifies the seat license.
7. General Ledger Reconciler — Accounting
Anthropic's job: Match transactions across systems, flag exceptions, draft journal entries for the controller.
Who does it today: Staff accountant, daily.
Open-source alternative: vas3k/TaxHacker — self-hosted AI accounting agent with receipt parsing and ledger automation. Smaller scope (SMB-grade, not Wall Street), but the architecture is similar.
Buy or build: SMB → TaxHacker. Mid-market+ → Anthropic. Above $100M revenue, the connector breadth (NetSuite, SAP, Workday) is a feature you don't want to recreate.
8. Month-End Closer — Controllership
Anthropic's job: Coordinate the multi-day close — accruals, reconciliations, intercompany, consolidation, variance commentary.
Who does it today: Controller's team, 5–7 days a month.
Open-source alternative: None at parity. Month-end close is heavily firm-specific and requires deep ERP integration.
Buy or build: Buy. The multi-step, multi-day, multi-system orchestration is exactly what Managed Agents are designed for.
9. Statement Auditor — Audit / Compliance
Anthropic's job: Tie out financial statements, check disclosures, flag risk areas. Output is an audit workpaper.
Who does it today: Big Four senior, weeks of work for a quarterly audit.
Open-source alternative: None public. Audit firms have proprietary in-house equivalents but the open-source ecosystem has not addressed this workflow.
Buy or build: Buy if you're not a Big Four firm. The Big Four already have proprietary versions. The buyer is the corporate side preparing for the audit.
10. KYC Screener — Compliance / Onboarding
Anthropic's job: Assemble entity files, review source documents, run sanctions / PEP / adverse-media checks, package escalations for compliance review.
Who does it today: KYC analyst team, 30–90 minutes per high-risk entity.
Open-source alternative: Some pieces (sanctions checking via OpenSanctions, adverse media via NewsAPI), but no end-to-end open-source agent.
Buy or build: Buy. The combination of KYC rules, sanctions data, and document parsing — packaged as a Cowork plugin with audit trail — is exactly what compliance officers want to put their name on.
The Framing Question: Sierra Multiple, or Anthropic Multiple?
The market context for this launch is Sierra at $15B — a vertical-AI customer-experience company at a 50–75× ARR multiple. That valuation is the proof point for "vertical agents are real, and the market will pay for the depth." Anthropic just played the same hand, in a different vertical, with one decisive structural advantage: they own the model.
💡 The structural difference: Sierra's moat is a CX vertical's depth on top of someone else's LLM. Anthropic's moat is the LLM plus the templates plus the connectors plus the Cowork harness. If you believe Sierra deserves a 50–75× multiple for vertical depth alone, then Anthropic at the same depth + model ownership + Wall Street's data partnerships is structurally underpriced.
The counter-read is real too: Wall Street has never met an AI vendor it didn't try to dis-intermediate within 18 months. Goldman, JPM, MS all have internal AI groups with mandates to ship the same thing in-house, and a perpetual political imperative to do so. Anthropic's pitch is that the templates are the price you pay to not hire that team — which works for mid-market firms where the team doesn't exist, and is a tougher sell at the bulge-bracket where it absolutely does.
What This Means for the AgentConn Directory
Two near-term updates we're shipping based on this launch:
-
Each Anthropic template gets its own AgentConn entry as it becomes generally available. Today's launch put them in private preview for design partners; we'll add directory entries when public access lands. Templates rated by category (
Codingfor Model Builder,Otherfor the finance-specific ones since AgentConn doesn't have a finance category yet — that's a schema gap we're addressing). - Open-source counterparts get cross-links. TradingAgents, Dexter, and TaxHacker are already in the directory; we'll add "compare with Anthropic Earnings Reviewer" cross-links so buyers see both options.
The honest answer for most buyers in 2026 is buy Anthropic's templates if you have a Wall Street workflow and budget; build on TradingAgents/Dexter if you have a quant team and IP requirements that prohibit data leaving your perimeter. The vertical agent wave is real — and as today's HN thread shows, both flavors (closed-source enterprise + open-source DIY) are converging on the same conclusion: the framework era is over, and domain depth is the only moat.
Further reading: Anthropic's launch announcement, The Decoder's IPO-runway framing, and Bloomberg's coverage. For the open-source companion, our vertical-agent-wave piece covers the GitHub-trending side. The architecture playbook is Anthropic's Managed Agents docs.
Originally published at AgentConn
Top comments (0)