How AI turns product managers back into builders

AI is changing how product teams build, learn, and ship. But “AI fluency” isn’t a status you read your way into. It’s something teams earn by building.Product thinker and Product Growth author Aakash Gupta lays out a practical pathway for PMs, designers, and engineers to level up. In his words, when leaders ask how to raise fluency, “the very first thing is how you behave as a leader. Are you openly trying the newest tools yourselves?”

He advises modeling the behavior you want to see: Try tools, implement them in your work, and show your team what you did. This is the essence of “the Builder’s Path.”

The Builder’s Path 

The builder’s path is the journey to AI fluency through making things. (By “AI fluency” we mean  the practical confidence to use AI to think, make, and decide, and to know when not to.) 

Teams often start with quick prototypes: chatbot mock-ups, automated research digests, feature specs drafted in minutes. The value lies in quickly surfacing what might work without committing real resources.

As prototypes pile up, teams start to stitch them into workflows. A rough experiment becomes a repeatable process that saves time every week. “That’s where the leverage shows up,” Aakash says. This is the murky in-between: not polished, but already changing how work gets done.

Over time, some workflows prove valuable enough to harden into code. That could mean a custom integration, an internal tool, or embedding AI into the product itself.

For leaders, the builder’s path is both a map and a mirror. It shows where the team stands today — dabbling in prototypes, relying on workflows, or ready to invest in code — and hints at the next step forward. 

Here’s a deeper dive into each stage. 

Stage 1: Prototype fast, learn faster

Aakash’s entry point is direct: “Usually people at this stage need to get their hands into building AI products, and I would start with prototyping.” He suggests beginning with a personal or internal use case, not production. Use approachable tools like Lovable, Bolt, or V0 to spin up something scrappy and useful.

The goal isn’t to build the next viral AI app. It’s to practice making things that are fast, small, and useful. He suggests starting with an internal pain point or a personal workflow you already understand.

Here are a few kinds of prototypes that help teams build fluency quickly:

• Meeting summaries that actually work. Use a no-code flow builder to summarize recurring project meetings and post highlights to a shared Confluence page or Slack channel. The prototype teaches you how to shape prompts, handle context, and manage quality.
• Auto-drafted support replies. Pull customer tickets tagged with a theme — say, “billing” or “setup” — and have an LLM draft first responses for human review. You’ll quickly learn what information the model needs and where judgment still matters.
• Lightweight research digests. Collect mentions of your product from community threads or social channels. Ask an AI tool to cluster the feedback into themes and generate a one-page summary. It’s messy at first, but it shows how AI can help teams listen at scale.

These experiments take hours, not weeks. The point isn’t polish, it’s progress. Each prototype surfaces what’s confusing, brittle, or unexpectedly helpful, and that’s where fluency starts to form.

Why this matters: Prototyping exposes knowledge gaps safely. As Aakash puts it, “step one is to know yourself. What am I most scared about as it relates to AI product building?” If your gap is concepts (transformers, RAG, agent architectures), tune your “content diet” to follow credible builders. If your gap is code, prototyping is how you close it.

Stage 2: From experiments to workflows

When a prototype proves useful, “go from prototype to a full workflow.” Aakash recommends chaining steps with no-/low-code agent builders (he mentions Lindy, n8n, Make.com, relay.app, Magic Patterns, among others). This is where leverage shows up: The same task now runs on a schedule, branches correctly, and interacts with the tools your team already uses.

Leaders can make this concrete by building and demoing three simple agents:

• Email triage and drafting. “Build an AI that drafts emails for you,” Aakash says. “Have the AI categorize what type of email: maybe it’s from someone in your org who needs coaching, or someone outside your org who needs routing. Then each of those categories has a system prompt that helps the model respond correctly.” As you refine the prompts, he explains, “you can actually build an AI agent that’s drafting 70–80% of your emails and cutting your email down 70–80%.”
• Slack summarization and response. “Do the same thing with Slack,” he says. Apply the same pattern — categorization, prompts, and iteration — to reduce notification noise and free up time for deeper work.
• Competitor insights. On a regular cadence, “build a system that’s giving you the right insights,” Aakash says. Have it collect recent launches from your competitive set, summarize the target users and features, and even “have the AI think about their strategy.”

This is still the “messy middle,” but it’s already changing how work gets done.

Stage 3: Harden into code

Next, translate proven workflows into durable systems. “The final step is to get it into the code: Cursor or Windsurf, one of these coding IDEs. Build your AI agent in code.” Aakash’s advice here is pragmatic: If code is your weakness, pull this step forward. Otherwise, follow the progression so you ship what’s already working (prototype, workflows, code).

For productionizing AI features, he highlights two core competencies:

• RAG done right. The hard part isn’t naming RAG; it’s data pipelines and retrieval quality.
• Evals. “Evals and RAG are two of the most important skills,” he says. “Write evals across the product to keep quality from drifting as prompts, data, or models change.”

What leaders must provide: access and guardrails

Aakash is blunt about organizational enablement:

• Access: Many companies “are not putting out the money for a no-code agent builder” or the API budgets those agents will consume. If you want fluency, fund experimentation.
• Governance: As agents proliferate, “you actually need governance… making sure that they’re not A) spending too much, and B) doing anything like giving access to systems that they shouldn’t have.” He suggests a formal competency within security focused on AI.

This combination — modeled behavior, visible examples, real access, and guardrails — turns pockets of experimentation into a repeatable practice.

Practice AI across the product lifecycle

Aakash maps 15 use cases where AI fits end-to-end so fluency isn’t confined to prototyping:

1. Using AI to monitor customer feedback on social channels.
2. Synthesizing Zendesk tickets.
3. Listening to Gong sales calls.
4. Analyzing community discussions.
5. Using AI for analytics signals (like conversion rates or expansion).
6. Generating interview guides.
7. Writing better survey questions.
8. Synthesizing qualitative research notes.
9. Drafting PRDs.
10. Doing impact sizing.
11. Prototyping with AI tools like V0 or Bolt.
12. Building production features with RAG.
13. Writing evals to measure LLM performance.
14. Using AI to analyze feature results.
15. Feeding learnings back into discovery.

His challenge to PMs: “I just walked through 15 use cases of AI. Ask yourself: ‘Which am I not doing? Which one am I gonna pick up first?’ Eventually, you’ll use all 15.”

A note on AI prototyping: the biggest shift in the PM job

Aakash calls AI prototyping “the biggest change that has happened to the product management job in the last few years.” Don’t just deliver PRDs and strategy docs — deliver AI prototypes alongside them. In his live demo, he parallel-prototypes with multiple tools, gives targeted feedback, and then downloads code to iterate in Cursor. The meta-lesson: Treat prototyping itself as a skill with technique, not a one-off trick.

Take your next step on the builder’s path

You don’t get AI fluency by skipping to the end. You build it, prototype by prototype, workflow by workflow, then code.

If you’re a leader, start by modeling the behavior and funding the sandbox. If you’re a builder, start small and ship something useful this week.

Aakash’s closing advice is a simple ritual that compounds: “Spend at least an hour a week trying out a new AI tool.”

Keep learning

Aakash’s framework appears alongside those of Kene Anoliefo, Laura Burkhauser, Ravi Mehta, and Elena Verna in Atlassian’s new ebook, AI fluency: The new product superpower. Read the ebook →And you can listen to his full interview with Atlassian here.