Project Managers Revolutionize Their Role with AI

The Evolution of the Project Manager Role with AI

For a long time, the role of a project manager primarily involved coordinating and aligning teams to achieve common goals. However, this traditional view is undergoing a radical change. Today, the most successful project managers in leading companies are not just managing teams. They are embracing advanced technological tools, including using real code for prototyping, querying data conversationally through language models, and overseeing AI agents specialized in coding. These new skills enable them to significantly enhance their efficiency and impact.

To support this transition, a new course has been developed by the author of this article in collaboration with Colin Matthews, a recognized educator and frequent collaborator. Titled Becoming a Native AI Builder, this program aims to teach the use of the latest AI tools, such as Codex, Claude Code, and Cursor, as well as various products from Product Pass. The course teaches how to use these tools to support discovery, create prototypes using a real codebase, ship changes to production via GitHub, and set up assessments to automate and improve work quality. The course is designed not only for project managers but also for designers, researchers, and anyone looking to integrate AI into their daily work. The first session starts on July 13, and discounts are offered to subscribers of the Lenny Newsletter. Annual subscribers receive a $600 discount, while Insiders get a $1,000 discount.

Hands-On Workshops and AI Expertise

In addition to the course, Colin Matthews is organizing a series of free workshops in collaboration with leaders from innovative companies like OpenAI, Cursor, Linear, Replit, and Lovable. These live workshops will provide participants with the opportunity to acquire and practice new AI skills alongside industry professionals. However, these sessions are exclusively accessible to paid subscribers of the Lenny Newsletter.

Colin Matthews is known for his pragmatic approach to teaching AI. With significant experience at leading companies such as OpenAI, Google, Stripe, Figma, and Microsoft, he has trained thousands of project managers in essential technical skills. In addition to his career as a product leader and founder, he has launched over ten SaaS products. Colin has also co-authored four guest articles with the author of this article, one of which became the third most popular article of all time. To celebrate the launch of the course, Colin wrote an article detailing the current possibilities offered by AI and how professionals can progress on the "leverage scales."

Understanding Personal Leverage Scale

The use of AI at work has become commonplace, particularly for writing documents, researching information, or creating small artifacts. However, it is crucial to understand the different levels of AI usage to assess where one stands and how to progress.

• Level 1: At this level, AI is primarily used to generate text, such as product requirement documents (PRDs), Jira tickets, or emails. Users then copy this text into other tools for further editing. Many professionals find themselves at this initial stage.

• Level 2: Here, AI is employed to create concrete artifacts, such as presentations, simple Excel templates, or small prototypes. AI no longer just produces text but generates tangible elements.

• Level 3: At this level, AI is integrated into external tools like Amplitude, Google Drive, Notion, and Canva to accomplish specific tasks. For example, AI can analyze customer support tickets or A/B test results, thereby replacing certain human tasks.

Practical Use Cases of AI

Initially, interaction with AI often involves simple exchanges. For instance, to draft a PRD, one might ask Claude to contribute to its writing. However, the AI has little context about the company or what constitutes a good PRD, requiring several iterations before achieving a satisfactory result, which will then be refined in Google Docs or Word.

The next step is to have the AI work more autonomously. For example, one might ask Claude to create a financial model comparing the cost of hosting an agent in-house versus using a managed service like Vercel. Here’s a prompt used recently:

Create a model that represents the costs if we build and host ourselves versus using managed agents. Research the engineering time saved and the computing costs of self-hosting versus managed. Look at other providers, such as Cloudflare, Vercel, or E2B that provide environments for agents for the price. Show both the pilot cost and the large-scale cost in the model, assuming we have 5 million agent instances running annually (where one agent instance is per hour).

The generated model can be viewed here. While revisions are often necessary, this approach represents an advancement over simple text copying.

The highest level of personal leverage is achieved when one can delegate entire tasks to AI. Take the example of a fictional product named Stride, similar to Strava. To analyze the retention of users sharing workouts with photos, one could connect a language model to an analytics software like PostHog to execute this task. Here’s a possible prompt:

Use PostHog to check if users who use the social sharing features have a higher 30-day retention than those who do not. Show me an HTML document as the final output visualizing the cohorts and any other useful data. Cite all your sources so I can validate.

The result is an HTML document detailing the retention data, with links to the sources in PostHog for validation.

Tips for Maximizing AI Usage

To enable AI models to perform tasks for you, it is essential to connect them to the products you frequently use via MCPs. Tools like Claude Code, Codex, and Cursor can connect to platforms such as Figma, Amplitude, PostHog, and Pendo. While this may seem complex, the process is relatively straightforward, and once set up, there’s no need to revisit it.

Once the connectors are in place, try using AI to accomplish common tasks, such as:

• Analyzing the success of a launch by reviewing recent customer tickets and online sentiment
• Checking the actual usage of a feature through product analytics events
• Summarizing a customer call recording and creating a prototype based on their feedback
• Updating your next sprint based on a change in roadmap priorities

Results may be disappointing at first, as the model still needs to learn how to meet your expectations. Continue refining the process until you achieve a good result, then formalize it by creating a dedicated skill in the same chat where the task was accomplished.

This workflow can be repeated to generate acceptable drafts for a variety of documents and projects, such as PRDs, roadmaps, marketing assets, survey analyses, prototypes, and Figma mockups. While the initial quality may not be perfect, it is possible to continuously improve it by continuing to work with AI.