I save the prompt that drafts our engagement letters and share it with my whole team. Is that an agent?

I set an AI to watch my inbox for receipts, drop them into a folder, and record the figures in a spreadsheet. Is that an agent?

I have AI that prepares our month-end, the whole thing, end to end. Is that an agent?

I don't think it matters much what we call an agent anymore. The thing is that even if you never actively create an agent in your business, the AI tools you are already using are becoming more and more agentic.

And you need to be… prepared.

From a chat to a team, one job

Let's follow one job along the line that is the agency continuum. A GST review.

Chat. I go in and ask about a transaction. Should it have GST, or not? It tells me, and I go and update it myself. This is the AI we have had for a couple of years now.

A saved skill. I have a great prompt, so I save it. Now I can hand it a GST report and it highlights anything that looks suspicious or inaccurate. Repeatable, every time. The way I like it. I share it with my team. Copilot calls these agents. Gemini calls them gems. Claude calls them skills.

Knowledge. Now I connect my AI to my systems, say, into my accounting app and my practice management system. Now it can see this client. Their industry, their size, the rules I saved as notes against them. So its comments on their GST treatment are specific to them, not generic.

Tools. Now I give it tools and let it actually do things. It puts together a workpaper for me and highlights items in red and green so I know what to look at. It writes notes inside my spreadsheet, or annotates a PDF.

Triggers. Now I don't even ask. It runs every night, looks for new transactions, pulls them from the accounting app, and leaves me a report of what it thinks I need to look at. On a schedule, or because something happened, a report landed, an amount went over a threshold.

AI employees. Now picture a team of these clever AI processes. One fetches the report. One gathers the client information. One compiles the workpaper. One chases the humans so they review and give their okay. And one checks the whole thing again at the end to make sure it all looks right.

This is the agency continuum and your tools are progressing along it whether you actively push them or not.

Where we actually are

Most firms I see today are working on saved skills and starting on knowledge. This is where I see AI reviewing compliance work, drafting engagement letters, creating AML risk assessment profiles and collating information for client onboarding.

Giving AI tools is the next horizon. It is still mostly with the early adopters. This is where risk spikes and we need powerful guardrails before we can comfortably sleep at night.

Triggers are popular for simple tasks: daily brief, inbox triage, weekly reports.

And AI employees are on the horizon. The innovators are playing there now. In their simplest form, they are AIs with knowledge and tools, but also a goal. They own a process or a target. So they get more autonomy and decision-making power.

What you’ll see in your AI platform

Even if you do nothing, agentic AI is coming to you.

Microsoft Copilot is the most conservative of them all, and it already has those saved-skill agents inside the basic chat. It has now introduced Cowork, with limited tools built in. It will rename your files, organise your inbox, create documents. But it will not send emails and it will not delete files.

The other day I had a few dozen documents to send to a few dozen people, the right one to each. Copilot Cowork drafted every one, with the right name and the right document attached to the right person, all sitting in my inbox waiting for me to review and hit send. It can build spreadsheets, copy information, download reports. So even if I never create an agent, my Copilot is becoming more agentic on its own.

What does this mean for us

Mostly that we need to educate ourselves, for two reasons. One, so we can start using these powerful tools to make magic and get the benefit. Two, this is how we keep safe. You cannot keep safe around something you do not understand.

So go and have a play with the agentic capabilities inside the tool you already use.

Tools like Microsoft Copilot and Google Gemini are built into your office environment and usually come with good guardrails, so people can only do what they should and their CEO can still sleep at night.

Tools like ChatGPT, and definitely Claude, are very, very powerful, but the blast zone is big too. There are far fewer guardrails on what you can let your staff do with them.

But we still need to understand them. This is the world we are living in.

Want to do this together?

Inside the AI with Inbal Club we are running Claude training in August, and we will repeat it monthly for as long as people want it.

At our in-person days in Sydney and Melbourne we will focus on Copilot, and bring in some Claude too, as a reference, not as the main tool.

Get in. There is so much magic to be had!'

-

Inbal Rodnay

Guiding Firms in AI Adoption and Automation

Keynote speaker | AI Workshops | Executive briefings | The Tech Savvy Firm

When you are ready, here is how Inbal can help:

From chatbot to AI employee in six steps

From a chat to a team, one job

Where we actually are

What you’ll see in your AI platform

What does this mean for us

Want to do this together?

© 2026 Inbal Rodnay. All rights reserved.

From chatbot to AI employee in six steps

From a chat to a team, one job

Where we actually are

What you’ll see in your AI platform

What does this mean for us

Want to do this together?

The magic I do with Claude (and why I'm still careful)

© 2026 Inbal Rodnay. All rights reserved.