Anthropic's New AI Agents for Finance

0:00Anthropic just dropped AI agents for financial services, and this is a news that no one's really talking about. It says today we're releasing at 10 ready-to-run agent templates for most time-consuming work in financial service. So, building pitchbooks, screening what KYC files, and closing books at the end of the month. So, each one ships as a plugin in Claude co-work and in Claude code. So, for managed agents. So, basically a team can put Claude on a real financial work in days rather than months. So, Claude now works across Microsoft Excel, PowerPoint, Word, Outlook coming soon, of course, through the Claude add-ins for Microsoft 365. So, new agent template for finance work. Each agent template is a reference

0:46architecture that packages three things. So, skills, instructions, and domain knowledge for the task, connectors, which is governed access to the data the task runs on, and sub-agents, additional Claude models that are called upon by the main agent for specific sub-tasks such as comparable selection or methodology checks. So, this is pretty amazing. You got research and client coverage, you got a pitch builder, a meeting preparer, earnings reviewer, model builder, and market researcher. Finance and operations, you get a valuation reviewer, a general ledger reconciler, and then month-to-end closer, statement auditor, and know-your-customer screener. Pretty cool. I'll put the link to this down below. You really need to check this out, especially work in the finance

1:31world. They basically released 10 ready-to-run AI agents in the finance space and for insurance. You know, we are talking things like pitchbooks and know-your-customer files and month-to-end closings, earning reviews, financial models, market research even. So, the reason why this matters is because finance is where fake agent demos basically go to die. We haven't seen much evolve in this space in the last year, But, Anthropic is shipping these features quickly. So, you can make a chatbot, right? Sounds smart in normal conversation. Maybe you give it some PDFs, some Excel spreadsheets, your QuickBooks. But, we haven't fully got the agentic workflow when it comes to finance. You know, you can even make a coding agent, build even a tiny app to

2:16trap some of your finances. You can make a research agent to summarize your website. But, finance in general is very difficult to build out because it deals with very sensitive information. So, if an agent helps build a pitch book or update a model or reconcile a ledger or a screen for know your customer documents, the work has to be actually traceable. Real quick, before we keep going, if you're watching this and you want to actually build with some of these tools, not just watch videos about them, then you're going to want to check out our community down below, Shipping School. We have a full Claude code course, a full Open Claude course, and four live bootcamps every single week where we actually help you get set up from scratch. Like, actually set this thing up, not just watch a tutorial and

3:01figure it out by yourself. And we also provide one-on-one coaching, so you could book a call with me, we could share screens, and I can help you get Claude code or Open Claude running on your machine. That's it, no fluff. I built this community because watching YouTube's only gets you so far. We launched it just 3 days ago and we have over 55 members. You need people around you who are actually building, people who hold you accountable, and coaches who can help you when you get stuck. I'll put the link in the description down below. Get in now before the price goes up. It has to connect to real data, it has to fit existing approval flows, and it has to work inside tools people already use. And at the end of the day, a human still needs to review it before it goes to a client or gets filed or

3:47triggers a real business action. That is why this announcement is pretty interesting and why we're even talking about it. Anthropic is not just saying Claude can help with finance. We already know that. They are packaging actual workflows that you could use today. According to the announcement, each template package, it packages three things: skills, connectors, and sub-agents. Skills are the instructions and the domain knowledge for that task, and connectors give govern access to that data. Sub-agents are extra Claude models the main agent can call for specific parts work, like comparable selection or methodology checks. This is the part builders should really pay attention to. The pattern is not prompt plus the tool. The pattern is packaged work. And I think that's where this whole agent space is going in general.

4:33We've seen it evolve crazy in the last few months. You know, it's not like, "Hey, here's an empty agent. Good luck." We're seeing packaged agents being sold as a service now. Agents with skills that work instantly the moment you install them. You know, it's more like, "Here is a packaged agent with an analyst workflow." Or here is a packaged agent that will handle your content. Here is a packaged agent that will handle your sales workflow. It already knows the task shape. It already knows the tools, and it already knows the approval points. Now, adapt it to your business. That is a better product than a blank chatbot, right? Come on. Let's walk through the templates together and see if they can be useful to you because they show exactly what Anthropic thinks companies will be paying for. The pitch builder creates target lists. It runs

5:20comparables and drafts pitch books for client meetings. This is a real investment banking work. You know, junior people spend tons of hours pulling data, formatting slides, and checking comps, and even turning analysis into a pitch deck. If Claude can do part of that reliably, the labor is going to be looking for work. I'm sorry. You know, meeting prepara also assembles client and counterparty briefs before calls. So, every business has a version of this. Before a call, somebody needs to know who the client is. What happened the last time they spoke, what changed, and what risks mattered, and what questions need to be asked in the upcoming meeting. Earnings Reviewer reads transcripts and filings, it updates models, and flags theses relevant changes. So, this is a very

6:05agent-shaped type of product, right? Earnings calls are structured, filings are structured, models have patterns. The analyst still needs judgment, but the first pass of read, extract, and compare, and flag is perfect agent work. Model Builder creates and maintains financial models from filings, data feeds, and analyst inputs. So, if you have worked in spreadsheets, you know the spreadsheet is not just math. It is assumptions, formulas, linked tabs, source data and version history, and team conventions. So, if Cloud is going to work in that environment, it has to understand the workbook and the business context, not just write a formula. Market Researcher tracks sector and issuer developments, reads news, filings, and broker research, and then

6:51it flags the items for credit and risk review. Again, this is not one question or one answer. This is ongoing monitoring. New information comes in, the agent checks the relevance, and then it summarizes it, and then it routes the risk flag, right? Then you get operations side. Valuation Reviewer checks valuations against comparables and methodology. General Ledger Reconciler reconciles accounts that runs net asset value calculations. So, month-end closer runs the close checklist, prepares the journal entries, and produces close reports. Statement Auditor reviews financial statements for consistency and audit readiness. Know Your Customer Screener assembles entity files, reviews source documents, and

7:36packages escalations for compliance review. That list matters because it is not just glamorous front office work. It is back office, middle office, compliance, and operations. The stuff every big company has to do even when nobody wants to talk about it. And that is where agents are going to make a ton of money. Not because they replace every person overnight, because they shrink the amount of human time required to get review. That phrase is important. Get to review. Basically, Anthropic is saying users stay in the loop reviewing, iterating on, and approving Claude work before it goes to a client, gets fired, or is acted on. That is right framing. A lot of people talk about agents like the goal is full autonomy now or tomorrow.

8:21But in serious business, the near-term win is not no humans. The win is humans stop doing the lowest value parts of the process. The agent gathers, the agent drafts, and the agent checks. Even the agent compares. The agent can even create a first version. Then the human reviews and fixes and approves and owns the decision. That is how adoption happens, and we're seeing it happen in real time. Another big piece to this puzzle is Microsoft 365. Claude now works across Excel, PowerPoint, Word, Outlook, and through add-ins, right? So, Outlook is coming soon, according to the announcement. The key phrase is that context carries automatically between applications. That's really nice. That means that work can start in Excel and then move into PowerPoint without

9:06re-explaining everything. That is actually bigger than it even sounds because business work does not happen in just one app. You're research in your browser, right? You pull your data from your vendor, you update a spreadsheet, you turn the numbers into slides, and then you write the memo. You can send the email, and you can jump on the call, then update the tracker. A normal chatbot breaks because it sits outside of all of that flow. You have to copy and paste context, then copy paste the results out, and it gets so old. Just right now, that's the old way to do it. So, if Claude can work inside of these apps like Excel, PowerPoint, Word, and Outlook, and then carry that context between the tools, it starts to look less like a chatbot and more like a co-worker with agentic workflows. The next big piece to this is deployment.

9:51Anthropic says these templates can run in two ways. One, as a plugin in Claude Co-work or Claude Code. >> If you're struggling to keep up with content, well, I'm about to save you about 40 days worth of work. I built something called Content Machine. It's 10 AI agents that run on the open Claude orchestration, and they handle everything. Scripts, thumbnails, X posts, blogs, outreach, clips, newsletters, all of it. So, I went from 1,000 subscribers to 4,000 subscribers on YouTube in 7 days using this exact system. Every single morning, I wake up and the content's already done. I spend maybe 15, 20 minutes reviewing and approving them, and I move on with my day. It works for any niche, fitness,

10:37finance, real estate, marketing, whatever you are building, and it is 100% completely customizable to your use case. So, you get the Mission Control dashboard, all of the cron jobs, everything I've built over the last 40 days helping me gain more and more people to subscribe and join the community. So, you plug in your own thing, and it molds it to you. It learns how you talk, and it writes so it doesn't sound like AI slop. $97 one-time. It's not a subscription. I'll put the link down below, and you'll thank me later. >> That means the agent runs alongside the analyst using the software already on their desktop. Two is as cookbooks for Claude managed agents. That means the same template can run autonomously on the Claude platform for work that spans

11:22a whole book of deals or a nightly schedule. The managed version includes long-running sessions, per tool permissions, managed credential vaults, and a full audit log in the Claude console where compliance and engineering teams can inspect every tool and call on that decision. That is enterprise checklist. Long-running sessions because real workflows do not always finish in one prompt. Permissions because not every tool should be available at all the time. And credential vaults because nobody wants random API keys pasted into prompts. Audit logs because if something matters, someone will ask what happened. This is where I think the agent market is splitting. Local agent systems like OpenClaw win on control, ownership, and customization, and running your own

12:07workflow across your own machine and channels. But hosted managed agent systems from companies like Anthropic win on convenience, governance, support, and approval inside of big companies. Both are real, they're just two different customers. For a solo founder, a local setup might be perfect. You can wire it into Telegram or Discord. You can wire it into your local database, your content pipeline, your GitHub repos, your calendar, and your weird scripts. For a giant financial institution, they may care more about managed credentials, audit logs, approval connectors, Microsoft add-ins, and vendor accountability. That does not make one right or one wrong. It just means that the market is maturing. Another important part is connectors. Anthropic says Claude connects to financial data and research platforms

12:53like FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, chronography, LSEG. They all announce connectors from Dun & Bradstreet, Fiscal AI, Financial Modeling Prep, Guidepoint, Intralinks, Third Bridge. Moody's also launched an MCP app that brings proprietary credit ratings and data on more than 600 million public and private companies in Claude for compliance, credit analysis, and business development. That is not just a partner list, that is a moat. Agents are only as good as the context they can reach. A finance agent without real data is basically a smart intern with no files. It can talk, but it cannot do the work. Give it governed access to the right systems, and now it has a real shot.

13:39Anthropic also points to Claude Opus 4.7 leading the Vowels AI finance agent benchmark at 64%. I always treat benchmarks carefully. A benchmark is not a real workflow. It is not your messy data. It is not your compliance policy, but it's still a signal nonetheless. Anthropic is clearly positioning Claude as a serious work model for high-value knowledge work, and they just increased usage limits with this new SpaceX deal just yesterday. So, they are seriously shipping and trying to buy as much compute as possible to keep up with the demand, which is a good sign. Finance templates is also just an amazing thing. You know, the Microsoft add-ons, the managed agents, the connectors, audit logs, and benchmarks all point to the same direction. And the lesson for

14:24builders is basically this. Most of us are not building hedge fund workflows, but the pattern of this workflow does transfer to us. A YouTube content agent is a template. A local business lead gen agent is a template. A reselling listing agent, that's another template. Another customer support triage agent is a template you could sell. A weekly finance tracker is a template you could sell. So, stop thinking of agents as generic chatbots. Start thinking them as repeatable work packages. What does the agent need to know? What tools does it need? What data does it need? What steps does it repeat? Where does the human approve? What output format does the business actually use? Where does the work get saved? This is the blueprint. And if you are building for clients, this is what matters even more. The

15:10client usually does not want an AI agent. They want the month-end report done faster. They want leads qualified. They want support tickets answered. They want invoices reconciled, and they want content drafted. They want the ugly work handled. So, sell the workflow, not the agent. That is what Anthropic is doing here. They are not saying Claude is smart, please buy it. They are saying, here are the expensive workflows for your company all that it already does, and here's a way to get Claude into those templates, plugins, connectors, and manage agents, and audit logs. That is a much stronger pitch. So, my take is this, Anthropic's finance agents are not just about finance, they are about a signal where the agent market is going. The winners will not just be the best models, the winners will be the systems that package real work, connect real

15:56data, fit inside these existing tools, and keep humans in the approval loop. For builders, this is amazing news. You do not need to compete with Anthropic on Wall Street. You need to copy the pattern at your level and apply it, and then sell it as an agent as a service. Pick one workflow, turn it into a repeatable agent package, give it the right context, the right tools, the right approvals, and the right output, and make it boringly useful. That is where the opportunity is, not another generic agent chatbot, and not another agent demo that looks for a cool 30-second workflow. You want real workflows with real data, real review, and real business value. That is why this announcement matters, and you need to be paying attention. That's why I make these videos. If you need help building out your first AI agent, or building anything like a web app, or a

16:41mobile app, or using Claude code, or getting Hermes agent installed, you got to check out the community. I just launched it about 40 days ago. We got nearly 200 members. You can check it out linked below. We have unlimited one-on-one calls, so if you need to share screens, we got one-on-one calls that are unlimited to you. I also do nine live calls a week in a group setting. We also have Claude code courses, and as well as open Claude courses in there. So, check it out linked down below, and we'll see you in the next one. Have a blessed day.