How to experiment with AI on your projects

In tech circles, we say that the latest wave of artificial intelligence (AI) began in 2012 and was kicked into overdrive at the end of 2022 with the launch of ChatGPT and the subsequent launch of other large language models (LLMs) like Microsoft Copilot, Claude and Gemini. They are the newest AI technology and, more importantly, hold the most promise for project management.

Where should you start?

The very short answer is that you should just do it. What matters is that you start the journey somewhere and that you iterate and keep trying as time goes on. That said, the best place to get started is with ChatGPT. Experimenting with Claude, Copilot or Gemini is also great, but if you’re going to try just one, ChatGPT is the place to start. It is the most popular and generally the most advanced.

If your organisation or situation allows you to, you can upload documents or images of all kinds to ChatGPT; otherwise, you can just ask generic questions or describe your situation without giving details.

Be as specific as you can

When using a search engine like Google, your goal is often to make your query as generic as possible to match other queries. When using an LLM, your goal should be to provide as much specific information or context as possible. As a concrete example, I have a snippet that I always copy/paste into ChatGPT before writing the rest of my chat: “I am working on a SaaS app called Dart, which does project management using AI.” Then, write the rest of your question, which can range from general to very strategic.

Here are some common queries and use patterns that I’ve compiled from talking with hundreds of project managers about how they use AI:

Brainstorm some solutions to a particular problem my users or customers are having.
Summarise this product feedback.
Write a spec for this project based on these user stories.
Break down this project spec into manageable chunks.
Prioritise, organise, group or categorise this work.
Advise me on how to communicate with and manage this stakeholder, including with role play.
Anticipate objections to this proposal so I can be prepared when they come up.

Find a way to give more context

As you can see, there are many ways to use an LLM, and they span the entire project life cycle. Many of these might require you to copy/paste or rewrite details from other places into the chat, but don’t stress too much about formatting. If you’ve never seriously tried using an LLM before to help with your standard work functions, this might seem a bit weird, but think of trying this out as an investment in skills that will become necessary over years to come.

When you get a result that you want to share with others, think of the AI output as a first draft and be sure to proofread it for inaccuracies.

If you get a result from ChatGPT and find yourself thinking, ‘Well, this is a good generic answer, but it is not very relevant to my particular situation because it is missing context from my organisation’, this is common but also easily solvable. Simply find a way to give the LLM more context. Consider what resources you would give an employee joining your team who was similarly still getting up to speed. You can talk them through where they went wrong or, if possible, provide some documentation so that you don’t have to spell it all out.

Any of this is possible with an LLM. Provide your onboarding documentation (for your organisation or project) and the LLM will have context to the extent that it is written down. If documentation is lacking or you prefer verbal communication, you still have an option: try ChatGPT voice or video chat and just talk it out.

Time to go deeper?

Once you have spent some time getting acquainted with the power of LLMs, there are a few ways to go deeper and use more tailor-fit tools. One way is to use features built into your LLM of choice to help it retain context about you. In ChatGPT, you can make a custom GPT, in Claude you can make a project and other tools have related features.

The underlying idea is to save context so that the AI can access what it needs to know and you don’t have to remind it details about your work every time you start a new conversation. One way I have seen this employed successfully by project managers I've worked with is to provide all the documentation about a given project, or portfolio of projects, and then directly converse with the now fully context-aware LLM.

Knowing your project portfolio will obviously improve its answers to both detail-based questions (‘For this next milestone, what is the most likely thing to slip?’) and higher-level ones (‘How can I ensure the success of this project?’), because it draws from historical source information.

In addition to the tools we’ve discussed, it’s entirely possible that the project management software that your organisation already uses has some built-in AI functionality. Tools like Jira, Asana and others have moved gradually in the direction of adding more LLM-enabled features. If you use Excel to plan projects (I’m sorry about that), you can use Copilot there. In general, it is worth checking your tool for built-in AI capabilities.

There are a lot of interesting upcoming tools, such as more autonomous programs called agents, that will keep the field moving forward quickly. These are very powerful technologies that will save you time and help you do better work, and I encourage you to start using them as soon as you can. Good luck!