AI July 2026

On 5 May 2026, Anthropic, the company behind Claude, released ten ready-to-run AI agents built for financial-services work, and made Claude available directly inside Microsoft Excel, PowerPoint and Word. The news was covered by the Australian Financial Review and the financial press, which tells you who it was aimed at: banks, funds and finance teams, not the average small business.

So why write about it here? Because the headline is finance, but the real story is the one underneath, and it applies to every business. This is the clearest sign yet that business AI is moving from a chatbot that answers questions to an agent that does the work, and the same Claude models behind these finance agents already sit inside Microsoft 365 Copilot, which your business can use today.

What Anthropic actually released

The release is ten agent templates aimed at the slow, repetitive parts of financial work. They fall into two groups. On the research and client-coverage side there is a Pitch Builder, a Meeting Preparer, an Earnings Reviewer, a Model Builder and a Market Researcher. On the finance and operations side there is a Valuation Reviewer, a General Ledger Reconciler, a Month-End Closer, a Statement Auditor and a KYC Screener that assembles client files and reviews documents.

The detail that matters for the rest of us is where they run. Claude now works as an add-in inside Microsoft Excel, PowerPoint and Word, with Outlook to follow, and it carries context across those apps. In Anthropic's own words, an analyst who has started a model in Excel does not need to re-explain it when the work moves to PowerPoint. Serious institutions are already named as users, including Citadel, BNY Mellon, Carlyle, Mizuho and Travelers. This is not a science project.

The real story: from chatbot to agent

Until recently, business AI meant a chatbot. You asked a question, it gave you an answer, and you did the work. An agent is different. It carries out a multi-step task across several tools, then hands the result back for a person to check. Read this set of filings, update the model, flag what changed. Run the month-end close checklist, prepare the entries, produce the report. Assemble the client file, review the documents, escalate anything that looks off.

Look at that second group of agents again, the General Ledger Reconciler, the Month-End Closer, the Statement Auditor, the KYC Screener. These are not exotic hedge-fund activities. They are the everyday admin of any business with a finance function: reconciliations, month-end, onboarding checks for new clients, reviewing documents for consistency. The work being automated for a global bank is the same shape as the work your bookkeeper or office manager does every month.

Why a small business should pay attention

Two reasons. First, the capability is arriving in the software you already use. Claude inside Excel and Word, and Claude already selectable inside Microsoft 365 Copilot, means the gap between what a global bank can do and what a twenty-person firm can do is closing faster than most owners realise. You will not buy a finance-agent template, but you will, before long, be able to ask the tools on your own desk to do more than answer questions.

Second, the workflows on display are a preview of what is coming to general business software. If a Month-End Closer can run a close for a fund, the same pattern is coming to the accounting and admin tools an Australian SMB runs on. The businesses that think about this early, calmly, will adopt it well. The ones that wait until staff are quietly using agents with no policy will adopt it badly.

The part worth copying is the governance, not the cleverness

Here is the most useful thing in the whole release, and it is easy to miss. It is not the agents. It is how Anthropic wrapped them. People stay in the loop, reviewing and approving the work. Each tool has scoped permissions, so an agent can only touch what it is allowed to touch. Credentials sit in a managed vault rather than being handed around. And every tool call and decision is written to a full audit log that someone can inspect afterwards.

That pattern, approved tools, scoped permissions, a human reviewing the output and a record of what the AI did, is exactly what any business needs before letting an agent near real work. An agent that can reconcile your ledger can also make a mess of it if it has the wrong permissions and no audit trail. Capability without governance is not an advantage, it is a liability. This is the heart of how we approach AgileAI: not switching on the cleverest tool, but governed adoption, which we wrote about in it is time to govern your team's AI use.

What to actually do about it

You do not need finance agents. You do need to be ready for the version of this that lands in your tools over the next year. Three practical steps.

  • Tidy your Microsoft 365 environment first. Agents act with the permissions of the person running them, and in most tenants those permissions have drifted wider than anyone has mapped. We covered the cleanup in preparing your Microsoft 365 permissions before a Copilot rollout, and it is the same groundwork an agent needs.
  • Put a simple AI policy and approved-tools list in place. Decide what staff can use, on what data, and where a human has to check the result. This is the core of AgileAI Foundations.
  • Start small and reviewable. Pick one contained workflow, keep a person approving every output, and expand only once you trust it. Even the best models are not perfect. Anthropic's own figures put Claude at the top of a finance benchmark while still scoring well under full marks, which is a useful reminder that these tools are powerful assistants, not unattended staff.

The shift from chatbot to agent is not hype, and the upside is real. It is also exactly the kind of change that rewards a business with its house in order and punishes one without. Getting the boring parts right, permissions, policy and a human in the loop, is what turns it from a risk into an advantage.

Frequently asked questions

What did Anthropic release for financial services?

On 5 May 2026 Anthropic released ten ready-to-run Claude agent templates for financial work, covering tasks like building pitchbooks, reviewing earnings, reconciling the general ledger, running a month-end close and screening clients for KYC. It also made Claude available directly inside Microsoft Excel, PowerPoint and Word, with Outlook to follow.

Can my small business use these finance agents?

The agent templates are aimed at finance teams and investment firms, so most small businesses will not use them directly. The part that matters for everyone is that the same Claude models now run inside Excel, Word and Microsoft 365 Copilot, so the capability is arriving in the software you already use.

What is the difference between an AI chatbot and an AI agent?

A chatbot answers a question and leaves the work to you. An agent carries out a multi-step task across tools, for example reading a set of documents, updating a spreadsheet and producing a report, while a person reviews and approves the result. The shift from answering to doing is the change worth watching.

Is it safe to let an AI agent do real work?

Only with the right guard-rails. Anthropic's release keeps a person reviewing and approving each step, limits what each tool can touch, vaults credentials and logs every action. That governance pattern, approved tools, scoped permissions, a human in the loop and an audit trail, is what any business should put in place before letting an agent touch real work.

Is Claude available in Microsoft 365?

Yes. Claude is available as an add-in inside Microsoft Excel, PowerPoint and Word, and Anthropic's Claude models are also selectable inside Microsoft 365 Copilot. For most businesses, Copilot is the simplest and most governed way to use Claude on day-to-day work.