AI assistants and AI agents in Microsoft Teams: conquering the office
Konnichiwa! Welcome to the Office Samurai podcast where we’re not just surviving the office we’re conquering it one automated process at a time. This time we’re going to be discussing AI assistants and AI agents. Those clever little AI creatures that are making our lives easier or at least more entertaining. I’m your host Andrzej Kinastowski one of the founders of Office Samurai a company that dares to ask “What if we did business consulting without all the bullshit?”. So whether you’re a business leader a tech geek or just someone who is tired of answering the question where do I find a procedure for that for the thousandth time you’re in the right place. Now grab your favorite katana or that emergency chocolate bar stashed in your drawer and let’s get to it.
The hype cycle: agent reality vs. pretenders
AI agents. Everybody wants to talk about AI agents. If you go to LinkedIn half of the posts are about AI agents. And you will see a lot of different people talking a lot of different things about AI agents. They will destroy us. They will save us. They will take our jobs. They will create amazing jobs. They will do everything perfectly or they screw everything up. Everybody talks about AI agents and everybody wants to do AI agents and a lot of people are kind of pretending to do AI agents.
Recently we got news about this company Nate. It was a company that claimed to have AI agents that would finish the buying process for you. So if you were on a e-commerce site and you wanted to buy something basically you would just use Nate to wrap everything up and they raised more than $50 million um in capital to expand the company. Recently we have learned that the automation rate of all of the shopping has been 0% and all of those transactions have been processed by hundreds of human contractors who are working in call centers in the Philippines and in Romania. There will be more and more of cases like this, because where there’s hype there’s people who want to use this hype to either make you successful or to screw with you. So you do need to watch out for bullshit around AI agents.
Very often what is presented as an AI agent is really a lot of ifs or just a lot of people doing the work which is funnily enough which is not that bad. You know when you think about all of the people fearing that AI will take their jobs here we have something completely opposite we have AI giving work to the people so you know from that perspective it’s not all that bad. And there are a lot of consultants um on the market who will advertise to you this idea that AI agents are already so good that you can build whole teams of AI agents that will be working together and they will have AI agent bosses and they will have AI agent auditors and basically AI agents all around.
Now this for now can work in a lab but it’s not something that you will see in your company this year or next year. I’m sure someday that’s how it’s going to be. But for now this is just testing This is all labs This is all uh working only in very very specific cases. But it doesn’t mean that it doesn’t make sense to go into AI agents It doesn’t mean that you should skip this technology. There is a lot of stuff that AI agents can already do for you. And I want to talk today about the things that we know because we have tried and we have implemented the things that already AI agents can do. The things that AI agents can already help you with.
Assistants vs. agents: defining the wording
Before we get into that we need to talk about the wording of it all. So AI agent is used to describe a lot of different things. And what we try to do in Office Samurai is to show people the difference between an AI Assistant and an AI agent.
What is an AI assistant?
Let’s start with AI assistants. What is an AI Assistant basically when OpenAI first showed their ChatGPT to the world all of the companies started thinking this is quite an amazing tool and it can talk to people um in a natural language. So how about we teach the model the knowledge from our company our procedures um our standards and we then have this agent that can help our people answering their questions about all of the knowledge of the company.
The RAG framework: retrieval augmented generation
Now teaching the model is not really the way to go because teaching a model is a very complex process and it’s quite costly. So you don’t really want to teach the model your data because that would cost you a lot and every time the data changes you would have to retrain it. That would not make sense Maybe in some very very specific cases. But what we can do is to use what is called a RAG which stands for retrieval augmented generation. And the idea is that you take the knowledge of your company for a given subject or for a given department and you prepare this data You slice it and and there’s a few other things that you need to do around it. And then you put it somewhere Usually that would be a vector database. A vector database is this magical kind of a database where we can store knowledge and where we can search for knowledge but not in a traditional way where we would compare words or phrases but in a way that we can look for meaning. So it doesn’t matter how you phrase your knowledge you can still find it in the vector database even if you put this knowledge into completely different words.
So in a RAG framework you put the knowledge in this vector database and then what happens is when the user asks a question to a RAG framework that is hidden behind some sort of a chat the RAG application first asks the database find me all pieces of data that could have something to do with an answer to this particular question that the user asked me. And the vector database gives us back chunks of the knowledge that may be in any way connected to the question that has been asked. And then the RAG can do some cleaning and so on on this data. But it basically sends this data as a context to the GenAI model telling it answer this question from a user in the context of this knowledge that I got from the vector database. And this way the model can answer questions in the context of our company’s knowledge without teaching the model itself.
So this is a really neat way of putting together an AI Assistant that knows your information but is way cheaper and simpler to build than your own AI model.
Case study: CyberOla assistant at Office Samurai
Let’s talk about an example In Office Samurai we have this AI Assistant called CyberOla. Ola is the head of our admin and HR. And we’re not a big company We’re only 40 people but our admin and HR already is getting a lot of repeating questions where people ask the same kind of questions over and over and answers to those questions they could find in our procedures and in our documentation but very often they just don’t even know where to search for it.
So we have created CyberOla as an alter ego of our Ola the head of the department and CyberOla is performing some of the functions of our HR and admin team. So every one of our employees can open a chat with CyberOla on Teams. This is a tool that we use for communication within the company. So they go on Teams and they ask CyberOla a question and they can ask different kinds of questions. It can be something simple like how do I book a meeting room and then CyberOla returns information on how to book a room and also along with this information it returns information about what is the document that it found this info in and what is the page number in this document.
We have put inside CyberOla 70 different documents HR and admin procedures internal regulations IT procedures new joiner packs. So all of the information that one of our employees could be asking about. We can also ask things like what are the benefits for our employees and then again CyberOla will find this information and it will give it to the user. It’s not just searching for information and copying pasting it it looks at the information and rephrases it.
You can go and ask it do I need to change my laptop’s name and then it finds information about the change of laptop’s name in our IT procedures and gives you information about yes you need to do it And this is how you do it. Even with the questions like “Can I bring my dog to work?”. In Office Samurai our employees are allowed to bring dogs to work We love dogs and we really like to have them around And so CyberOla will tell us “Yes you can bring your dog to work”. But the interesting thing is we do not have a policy for any other pets just for dogs. So if you ask it can I bring my parrot to work it’s actually going to tell you I do not know I do not have information about this in my knowledge and it’s going to tell you please contact the real Ola the human Ola and she will give you the answer. And this to our surprise was one of the hardest things to achieve in our GenAI Assistant and this is because generative AI models they do tend to hallucinate a lot.
We have been playing with some AI agent builders a few different ones and there’s a lot of AI agent builders out there that allow you to build your own AI Assistants very easily. The problem that we had with basically most of those off-the-shelf solutions is that when we asked it can I bring my parrot to work very often they would say yes you can bring your parrot to work even though we have no information about bringing parrots to the office. If you use one of those off-the-shelf solutions… you really have to be careful about hallucinations because if you have one of those easy to use tools it’s quite hard to adjust them exactly to your needs and you do not have as much control over how it works compared to a custom automated tailor-made solution like CyberOla.
Opportunities for assistants across the organization
Now that you know what AI assistants can do take a moment to think how you could use them in your organization because there’s a lot and a lot of opportunities. Think about this You have your sales team and your sales team could have an assistant that gives them information about your products about specifications is able to compare things. Or you can prepare one for your finance team you feed this assistant all the rules and practices and procedures for your finance and then if they don’t remember how to book something or who can do this particular thing they can go to it and it will tell them. You could even do one for your IT service desk you know so they don’t have to ask users to restart the computer over and over An AI Assistant can do it for them. So those are AI assistants and I believe they have huge potential in our business processes and in the automation of the tasks that we and our people are doing.
Agentic tools in action: automating HR tasks
But there is the next level to it And the next level are AI agents. And we want to separate AI Assistants from AI agents because AI agents if you say AI agent this implies that this AI has actual agency. so from our perspective if it only answers questions it’s an AI Assistant If it can also do things for you then it is an AI agent.
So we have implemented certain things into our CyberOla to make it agentic. One of the things that we do as a company we provide a lot of trainings. So when one of our trainers provides a training every time they need a training survey. So now a trainer can go to CyberOla and say hey I need a survey for the training I’m doing and the framework that we have created it understands that this is not a knowledge kind of question so it doesn’t go to the RAG framework and to the logic of an AI Assistant it understands that this is where we should use one of the tools that CyberOla has.
When I give this information to it it contacts through an API First it contacts JotForm which is a tool that we use for our surveys and through an API it creates a standard survey for a training. After it has created the survey through another API it creates a short link to this survey and also generates a QR code that the trainer can just copy and paste to their presentation. So now our admin team doesn’t have to do this anymore Our trainers go to CyberOla and they create the surveys for themselves using this AI agent.
Another thing it can do is in the Polish labor law when you buy glasses your employer needs to reimburse you at least for the part of the cost. And to do this you need a form that you fill in and you send it to your admin team. We have created another agentic tool where if you say to CyberOla hey I need to reimburse my glasses it’s going to ask you all the relevant information that it needs to create this form for you. Once I give this information to our CyberOla it creates a PDF with all of this data I have provided injected in the right places and it uploads it to the channel on which we are speaking.
And we have other agentic tools implemented into our CyberOla. It helps us find booking slots for people within the company by contacting the Microsoft Graph API. It can prepare other HR documents and we keep on adding new functionalities to it.
The future is agentic: endless possibilities
And so if you think about those AI agents you can again think about how they could help your teams. When I talked about an AI Assistant for your sales team you could add to it agentic automation. It could also update the status of their leads in your CRM. It could be putting notes that you dictate to it about the meetings you took It could prepare quotes and offers for you.
For your finance team it could not only tell you how to do things and guide you to the procedures that you need but your people could be requesting an agent to do a booking to do a clearing to prepare a report. For your IT service desk you could have an agent that allows a lot of self-service for your people. So the possibilities here are basically endless.
Conclusion: AI will be business as usual
And the more we think and the more we work with AI Assistants and AI agents the more we see that this is something that could in time it’s going to take time but in time it could fundamentally change the way organizations operate. Today robotic processes automation is business as usual. And I strongly believe that in a few years AI Assistants and AI agents will be business as usual.
All in all Office Samurai’s perspective on AI Assistants and AI agents is the technology is now ready the possibilities huge and now it’s the high time to get it implemented. you’ve reached the end of another Office Samurai podcast episode. Big Arigato for listening. Massive thanks as always to the irreplaceable human element behind the scenes and Anna Cubal our tireless producer. Recorded as always at the legendary Wodzu Beats Studio. Until next time keep a firewall strong and your AI agents on a short digital leash. Mata ne.
