AI Agents Are Coming! Let’s Make Sure You’re Prepared
We are hearing a lot about AI agents. It’s no longer a future thing. It’s a now thing.
Let me show you what they are so that you become a smarter builder and a smarter buyer.
Having a Look Under the Hood
I often like to understand things from the inside. I lift the hood, I have a quick look, and I go, OK, I get it. I understand what we’re talking about. So, let’s do that.
The Simplest Way to Get Started
There are many platforms for creating AI agents. For the purpose of easy understanding, let's look at the simplest one that has the most integrations and is the easiest to use.
The easiest one is Zapier Agents.
When we create a new agent, we see a list of templates. That makes it easy.
In the video, I go into the inbox categoriser for Gmail.
And there’s a prompt there:
When you ask me to categorise emails, I’ll do the following. Retrieve all emails in your inbox. For each email, analyse its content, determine the category, important, work, personal or to do, and then I’m going to apply the selected label and provide a justification.
This is the heart of an agent: a prompt that tells it who it is and what it should do.
What Makes It an Agent
Different to the prompts we’re used to seeing with AI, agents have tools.
In our case, it has the tool to find emails in Gmail and to add a label to an email. Other tools can allow it to visit a website, search the internet, or link to thousands of systems to fetch or store or update information.
That’s what makes an agent: a well written prompt with tools.
The agent has an agency through its tools and autonomy to decide when and how it’s going to use these tools.
It’s not a deterministic flow chart. It has tools. It will do its best with them.
Agents Have Triggers
Triggers spring our agents into action when the time is right.
Do you want this agent to work on demand when a user is asking? Do you want to schedule it every day at 4:00 PM? Do you want to trigger it as part of another workflow?
Or do you want a trigger by an event in another app — for example, when a new spreadsheet row is created, a new event is created, or a calendar event is started or ends?
We set up triggers as part of the process of creating the agent.
How It Looks in Microsoft Copilot Studio
In the video I also show what it looks like in Microsoft Copilot Studio. Similar to Zapier (and all the others) the agent has instructions, which is its prompt.
It has knowledge, which can include files, templates, policies, and any background knowledge the agent might need.
It has tools. You can even give it other agents as tools so they can complete parts of the process together.
With Microsoft Copilot, you can publish the agent into Teams, SharePoint, or into Copilot environment itself for the rest of your team to use.
Using Agent Mode in ChatGPT
ChatGPT has just released its Agent mode, so it can start doing things for you, not just talking about things.
The ChatGPT agent mode has its own browser.
It can go and click things, add items to shopping lists, make purchases on your behalf, complete online forms, create files, any really do anything you can do in a browser.
The ChatGPT agent is still a baby. It’s a bit slow, it’s a bit clunky. It can work on something for 40 minutes and then just stop for no reason.
But that’s OK. It’s learning. It will get there and it will get there fast.
What You Should Do Now
Agents are where real automation with AI is going to happen.
We are going to interact with them in two ways:
We’re going to build simple ones like we’ve just seen.
We’re going to buy or rent ready-made agents that others have developed.
You should start playing with agents. Probably start with an easy environment like Zapier because it has the easiest user interface and the most integrations. So you are the most likely to be successful here.
Even if you’re not going to build agents long term, it’s important to have a play with them now so that you have a real understanding of what they are and how they work.
Then, when someone comes and offers to sell you an AI agent, you'll know what you’re looking at. You can tell if it’s just a glorified prompt. You can ask questions about how it’s built, what it’s built on, and how it’s going to work for you.
Think Carefully About Security
When we deploy an agent and let people interact with it, staff or external users, they now have access to an AI tool that has access to our systems, and autonomy to do things.
And we know AI. AI can have a mind of its own. It can be convinced to do things it’s not meant to do. It can be tricked. It can be pushed. It can be confused.
We need to think very carefully about security and how we make those agents secure.
For now, I don’t think we should expose any agents to the world to play with until we figure out how we’re going to keep them secure.
The whole world of security around AI agents is just developing. Experts are identifying vulnerabilities and developing the protections around them right now.
But we’re not quite there yet.
Live Agent Interview This Friday
It’s a really an interesting space to watch.
In our AI bootcamp this Friday, I’m bringing in a human developer, software developer that builds AI agents, and I’m going to interview them in front of everyone as if they’re selling me the development project of an AI agent.
I’m going to ask them all the questions to make sure that this agent is safe and is protecting me.
If you want to join, register here
—-
Inbal Rodnay
Guiding Firms in Adopting AI and Automation
Keynote speaker | AI Workshops | Executive briefings | Consulting CIO
Want to receive these updates straight to your inbox? Click here: www.inbal.com.au/join
When you are ready, here is how Inbal can help:
For CEOs, partners and business leaders. Everything you need to know about AI without the noise. Inbal shares the state of AI, recommends tools, and answers your questions about strategy, implementation and safe use.
Only what's real, no hype, no noise.
This is a one-off session for your entire leadership team.