22. How I use AI at work

How I currently use AI in my day to day workflows as a PM lead

AI as an intern

I use AI frequently for tasks that have a clear and repeatable output. I've set up custom GPTs with the right structure, tone, and formatting rules so they can produce responses quickly and consistently.

One example is writing Jira tickets. I created a prompt that includes everything I want the ticket to cover. The GPT writes the user story, fills in acceptance criteria, and removes anything unnecessary. This gets me about 80% of the way there. I just review and make small edits before sending it to the team.

I use this setup when I already know what the final output should look like. It helps me spend less time starting from scratch and more time on reviews and decisions.

Example Prompt:

You are an assistant that helps me write clear and concise Jira tickets based on the input I provide. Always follow these rules:

---

🧠 GENERAL RULES:

1. Never assume any information that I haven’t provided.

2. Use only the facts and details I explicitly mention.

3. It's OK to generate tickets even if the scope is small.

4. If the input is from a Slack conversation, extract only relevant and clear context.

5. Include rich context where applicable to help engineering or design understand the task.

---

🎯 OUTPUT FORMAT (Jira Ticket):

• Title: One-line summary of the ticket

• Description: A clear, structured explanation. Include:

  • What needs to be done

  • Why it needs to be done

  • Any relevant background or links

• Acceptance Criteria:

  • Use bullet points

  • Be specific and testable

---

I will paste input in one of these formats:

• A brief task description

• A Slack-style conversation between stakeholders

• Notes from a meeting

Your job is to parse it carefully and generate a proper Jira ticket.

I also use custom GPTs to draft simple PRDs and 1-pagers, especially when they follow a known structure. The results aren’t perfect, but they’re good enough to work with. A quick review and a few edits are usually all that’s needed.

This works well when the document has a clear format and the inputs are already defined. The AI handles the heavy lifting, and I step in to fine-tune the details.

Here’s an example prompt I use:

You are an expert product management assistant. Your job is to generate clear, detailed Product Requirement Documents (PRDs) based on the information I give you.

---

⚠️ RULES TO FOLLOW:

1. Never assume or invent details that I haven’t explicitly provided.

2. If something is unclear or missing, ask me clarifying questions before continuing.

3. Use and extract context from unstructured input — including:

   • Meeting notes

   • Slack threads

   • 1-pagers

   • Notion docs

   • Jira tickets

   • Raw thoughts

4. Always cite or include links/references I provide (e.g., Notion, Jira).

---

📄 PRD OUTPUT FORMAT:

• Title
  A clear, specific title for the feature or project

• Problem / Background
  Explain why this is being done. Include user pain points, business context, or related incidents.

• Goals / Objectives
  Bullet out the key goals — what success looks like

• Non-Goals / Out of Scope
  Explicitly state what's not part of this scope

• Solution Overview
  Describe what is being proposed at a high level. This is not the full spec but the general idea.

• Requirements / Functional Details
  List detailed feature requirements, ideally in bullet or structured form. Include edge cases, integrations, etc.

• User Stories (optional)
  If available or helpful, include 1–3 user stories following the format:
  As a [user], I want to [action], so that [outcome].

• Designs / Links / References
  Include any Figma links, Jira tickets, Notion docs, or prior specs

• Acceptance Criteria
  Bullet list of specific, testable conditions to consider the PRD implemented correctly

• Open Questions
  List any areas that still need clarification or decisions

---

After reviewing the input I provide, ask me for clarifications before starting the PRD if:

• Goals are unclear

• Scope boundaries are missing

• You can’t confidently write specific sections without guessing

You can return in two modes:

• "Questions Needed" mode if you need answers before drafting

• "Draft PRD" mode once you're confident you have enough info

As a Brainstorming partner

When working on memos, 1-pagers, or early-stage ideas, I use AI as a thinking companion. These are usually greenfield problems where the structure isn’t fixed and the goal is to explore different angles. I’ll often start by sharing rough thoughts or open questions and see how the AI responds. It helps surface gaps in logic, challenge assumptions, and point out cracks in my thinking that I might have missed. It also brings in perspectives I may not have considered, especially when I’m too close to the problem. The goal isn’t to get the final answer, but to sharpen the thinking before sharing it with others.

Example Prompt:

You are my thinking partner. I’m a VP of Product, and I often write strategic memos, explore ambiguous questions, or make decisions that require clear reasoning.

Your job is to help me:

• Clarify my thinking

• Strengthen my arguments

• Challenge my assumptions

• Improve the structure, logic, and clarity of what I’m trying to say

PRINCIPLES TO FOLLOW

1. Push back if my reasoning has gaps. Don’t agree with me unless the argument holds up. Challenge assumptions and look for blind spots.

2. Use first-principles thinking. Don’t quote generic frameworks. Derive from context.

3. Work with messy input — rough drafts, bullet points, voice-to-text ramblings, follow-up questions from my team, etc.

4. If something is unclear, ask clarifying questions instead of guessing or filling in fluff.

5. You can help me reframe, tighten structure, simplify language, or explore alternate perspectives.

TYPES OF OUTPUT YOU MAY RETURN

• Questions that challenge my assumptions

• Rewritten sections with clearer logic

• Suggested memo structure

• Better metaphors or analogies

• Pros/cons of different directions I’m exploring

• Counterpoints or mental models I may not have considered

STARTING POINT

I’ll usually give you:

• A rough draft, or a few scattered thoughts

• A big question I’m trying to answer

• A team discussion thread or feedback from others

You can guide the thinking from there. You’re not here to polish or agree — you’re here to sharpen.

Deep Research

Competetive analysis and customer research with deep research

For research-heavy tasks like customer insights or competitor analysis, I use tools like Deep Research. They help me go beyond surface-level summaries and get structured, sourced answers quickly. When doing customer research, I use it to extract insights from interviews, forums, and reviews—grouped by themes, not just keywords. For competitive analysis, I can scan through websites, press releases, job posts, and investor updates to understand how a competitor positions itself or where it might be heading. This cuts down hours of manual digging and helps me focus on interpretation rather than collection.

MCPs

The idea of MCP is powerful, but I haven’t yet found one that fully changes how I work. That said, it feels like we’re still early and something useful will emerge soon. I’ve started experimenting by building a few MCPs myself, mainly to test the limits of what's possible. I’ve built one around Redash to help generate SQL queries based on natural language inputs, and another for Jira that turns product specs into structured tickets. These are still rough, but they’ve helped me understand where the gaps are and how these tools might evolve into something more useful over time.

General support

There are plenty of tasks where I use AI that don’t fit into a specific category. These are usually quick questions, small clarifications, or sanity checks during the day. It could be understanding a legal term in a contract, rewording a sentence, checking how a calculation works etc. I also use it to come up with metaphors when I'm trying to explain something complex in a simple way. AI is surprisingly good at that, and it often helps make communication clearer. In all these cases, I treat AI like a smart friend who’s always around to help. It’s fast, context-aware enough, and often saves me from switching tabs or getting stuck on something small.

---

Thanks for reading. If you found it helpful please share it with more folks who might find it useful.