Getting your Project Management Office ready for AI Agents
In July 2023 I wrote an article on how to get your PMO ready for AI. A lot has happened since 2023, and a lot of expected changes have not happened. Most PMO’s still need to get ready for AI, but AI has also advanced somewhat. Therefore, this article will get you up to speed with what you should do considering the latest advancements in AI. You can read the original article here. This article is a refresh of the original, bringing it up to date.
What’s new since 2023?
In AI terms a week is a long time. However, while many new AI refinements have come out there is not necessarily a large change in how PMO’s are operating as a result. Generative AI became popular and still seems to be looking for a killer use case in the PMO. And that statement actually is what is wrong. People are looking for individual solutions within the PMO, instead of doing the one thing they should have been doing since their PMO was formed: clarify your strategic reason for existing and look at how to improve your processes and ability to deliver on your charter.
And that is where the one big thing since 2023 does emerge: AI agents.
What are agents and why should I care?
An agent is a tailored knowledge workflow, that brings AI and knowledge into processes/workflows so that the agent can refer to many information sources and make decisions. An agent can be seen as an entity that is autonomous, able to take in data from many sources, use AI to read a set of instructions, and then reason over the data and work out what to do next, invoking one or more actions such as doing something, updating a system, sending an email, etc.
​This is different from a traditional workflow or chatbot, as they lack reasoning.​ An agent is more than just adding AI to a workflow. An agent works with other agents to become an AI powered workflow.
The reason you should care is because unlike an AI chatbot which is just an assistant to you if you ask it to do something, AI agents are part of your process. Agents automate parts of the process and can act on your behalf as well. For example, a project may be requested. A governance agent could check that it is in line with strategy for you. But there is some preparation to do to take advantage of agents and you need to prepare as Microsoft’s agents come with Microsoft Copilot which you may already have.
What about Microsoft Copilot?
Microsoft is driving agents heavily as are others. Microsoft Copilot can be seen as the user interface to AI, and the front end to agents. Copilot gives you a natural language way to describe what you need and then calls agents to get that done for you. And many organisations have a form of Copilot already. But the thing to remember about Copilot is that you must do something to prepare for it. On its own it will be useful, but if you follow the steps below you will be able to extract value from it. Otherwise, it’s like being given a car but not being taught how to drive.
It’s not automation
Before we get into the details about preparing for AI, a little more positioning of agents. Above, and generally, people will describe agents as a form of automation. That is only very loosely true. An agent is not automation as automation works to set rules and is predictable. An automated workflow always works the same way and has no autonomy. An agent has a set of instructions and autonomy to work out what to do, just as a person does.
This means that as you prepare you need to:
- Create instructions. Later I will point out that an agent is not a person but sometimes gets described as a role in the PMO. If you had a new hire, you would onboard them, so in the same way you need to create documented instructions on what you want your agent to do.
- Put guard rails in place. Use agents to check on agents, or people to check. Just like governance puts boundaries around autonomy of people.
- Keep people in place. People are part of the process and should use agents as they work through the process, instead of the entire process being automated.
And it’s not a person
Above I said an agent was like a person. And the view of an agent as a role is a good way to explain it to beginners. But it isn’t quite true. Agents have less autonomy than people. You will have a lot more agents that you will people. Agents can’t learn by themselves. Don’t think about agents as roles. Think of agents as a Standard Operating Procedure (SOP). One person can handle several SOP’s, but agents are best when they handle one only.
This means to prepare you need to:
- Map your process so that you can identify the SOP’s, as these will form the input to your agents.
- Put people in place to trigger the agents, by getting it to execute SOP’s. You wouldn’t leave a junior alone without instructions, so don’t leave an agent to work by itself.
Start from established processes
Don’t try to automate something that isn’t currently well understood or consistent. And don’t add AI to it. Turning it into an agent won’t suddenly solve the issues you have with a lack of process. So, to prepare:
- Get agreement on a consistent process, and make sure it is mature.
- Document the process end to end.
- Work out who does what, and what SOP’s could be turned into an agent to help them.
Prepare the people you do have
You want to keep humans in the loop. So, prepare the people:
- You need to include them in the planning to identify the SOP’s that they will need an agent to perform.
- And you want to provide training and familiarisation to them. Sensei can help you with this.
- Take a change management approach as there is a lot of uncertainty around AI, so engage your change people in advance.
Look at your entire process
Try not to focus on individual use cases. Map out your process and identify where your pain points are. Where can you benefit from:
- Saving time.
- Accessing more complete data and information.
- Support in making decisions.
- Automation of repetitive tasks.
Make available and prepare your data
One of the most consistently repeated messages we give to clients is to prepare your data. Think about what AI uses: the data you give it. Imagine if you had a new starter in the PMO who was given a mismatch of old project documents and told to use those to figure out the best way to handle the next project. What would the outcome be? Therefore, to prepare you need to:
- Think about what data you will need and clean it. You don’t need all the data, you may need a subset. Typically for a PMO consider:
- Business cases. These are a great source of information about what has been done in the past.
- Lessons, again a great source of information on what not to do in the future.
- Risks and issues, which help educate people on what can go wrong.
- The rules and regulations, and how you operate. Not all of that may be written down.
- Clean project documentation describing what was delivered, so that it can be used again in future.
- Think about what examples you would provide to a new starter and prepare them for your agents.
- Store them somewhere a future agent can find them.
- Think about the future data sources such as your project solution that you want the agent to connect to, and make sure you can get data from them. With Altus and Microsoft Copilot this is taken care of for you.
Classify your data
As part of preparing your data you should also think about how to classify your data. How would AI identify a business case, or a status report? This really should be something you prepare with your IT team about data overall and is often done as part of a SharePoint implementation but may not be up to date anymore.
Secure your data
Tools such as Microsoft Copilot will respect the security you already have in place. If you can’t see a document in SharePoint now you won’t be able to find it with Copilot. But not everyone has security setup well on their documents and data. Therefore, you need to:
- Ensure your security is defined on your source data. IT can help you with this.
- Understand any compliance rules that will apply and make sure any tools are compliant. An advantage of the Microsoft AI tools is that the data stays with you under your control.
Set up a feedback process
Creating agents or Copilots or any AI is an iterative process. You will need to tweak and adjust it. So, leave time for that, and make sure you have a process in place where you can review the results. And then actually update the AI. Don’t leave it and then work offline instead! So, make sure you budget for ongoing tweaking of the agent.
And finally, don’t worry about what AI can’t do yet
AI is growing all the time. The quickest way to get value is the safest. Start simple with one basic use case, don’t try to do it all at once. AI works best as a simple proof of concept approach to get started and then iterate.
If you’d like to discuss how to get your PMO started on the AI journey, we are happy to have a chat.
