How I AI“I’m incapable of doing my job without AI”: How this PM uses Claude + ChatGPT as his second brain
CHAPTERS
Why this PM is “incapable” without AI: outsourcing context switching
Claire and Amir set the premise: PM work is constant context switching across meetings, initiatives, and stakeholders. Amir explains how he offloads that cognitive burden into siloed AI “brains” in Claude and ChatGPT to keep momentum and decision quality high.
- •PM pain: juggling many parallel threads and rapid context switching
- •Idea: create separate AI “brains” per initiative to retain context
- •Result: faster execution and less mental load; increased dependence on AI
Inside a GPT/Claude “project brain”: files, instructions, and persistent threads
Amir demos what a typical project looks like: a growing set of uploaded files plus explicit instructions, alongside long-running conversation threads. The structure is designed to accumulate context over time rather than restarting from scratch.
- •Projects contain: uploaded source files + custom instructions
- •Brains start small (2–4 files) and expand (20+ files) over time
- •Threads act as an ongoing working memory for the initiative
- •Goal: keep context localized so work feels “intuitive and fluid”
What to upload first: kickoffs, PRDs, and any “ping-pong” reference material
Amir explains his starting kit for a PM knowledge base: anything that defines the situation and goals (kickoff decks, PRDs, docs). He calls the early phase “ping pong”—feeding references so the AI understands what success looks like and can collaborate effectively.
- •Start with whatever anchors the initiative: decks, PRDs, docs
- •Upload “any data whatsoever” to establish shared context
- •Use the first thread to align on goals and the problem space
- •Treat the AI like an onboarding teammate that needs references
Finding unbiased customer truth: why Amir went to Reddit for AI agents signals
Leading an AI agents initiative at monday.com, Amir needed external, non-company narratives about what people expect from “agents.” He sought broad, unbiased conversations to ground internal hype and align priorities with real user expectations.
- •AI agents: high hype, unclear definitions, many internal opinions
- •Need: outside-in signal to balance leadership visions and assumptions
- •Reddit chosen as a rich, candid source of user discussions
- •Intent: understand expectations, desires, and real pain points
Claude as a technical copilot: step-by-step building a Reddit scraper
Amir walks through using Claude to go from an ambitious prompt (“scrape everything said about monday.com online”) to a practical plan. Claude narrows constraints (APIs, paywalls), then provides a beginner-friendly setup guide, environment steps, and a working script.
- •Start broad, then constrain to accessible data sources
- •Claude identifies API limitations for LinkedIn/Twitter vs. Reddit
- •Hand-holding setup: terminal, Homebrew, packages, Reddit dev account
- •Claude generates the script and explains where to insert credentials
From scrape to ‘monstrous’ dataset: producing 34,000 rows (plus competitor angles)
The scraper outputs a large CSV of Reddit conversations (~34k rows) tied to Amir’s themes (agents/AI/monday.com). He extends the approach to competitor comparisons and additional files to capture a wider market narrative.
- •Output: large CSV of themed Reddit threads (~34,000 rows)
- •Extends queries to competitor-related discussions and comparisons
- •Creates multiple files to segment different investigative angles
- •Turns qualitative chatter into a reusable dataset
Claude as analyst: frequency tables, prioritization weights, and quote-based validation
Amir returns to Claude with the dataset and asks for an analyst-style breakdown: topics, frequencies, and weighted prioritization. To reduce hallucination risk, he spot-checks via keyword searches and requests direct quotes as evidence to verify themes.
- •Ask for structured synthesis: topic table, % frequency, weights
- •Use numbers to communicate priorities back to team/leadership
- •Spot-check with keyword searches rather than reading everything
- •Require quotes/citations to validate summaries against the raw file
Turning raw research into a persistent brain: uploading Reddit CSVs + internal context
Amir shows how the scraped Reddit files become part of the project’s knowledge base alongside internal kickoff materials and company/product references. This creates a single AI workspace that understands both monday.com context and external market expectations.
- •Upload Reddit CSVs and analysis into the same project repository
- •Add internal kickoff deck and product references for grounding
- •Use PDFs of internal/external material to standardize ingestion
- •Outcome: AI can reason with both internal goals and external signals
Knowledge base mechanics: ‘everything is text’ + strong instructions and pushback behavior
They discuss practical KB-building tricks: printing webpages or slides as PDFs and uploading them as scoped source-of-truth artifacts. Amir also emphasizes instruction design—he wants the AI to be candid, challenge ideas, and avoid overly supportive agreement.
- •Convert slides/web pages/support/pricing pages into PDFs for upload
- •Scope content intentionally (e.g., positioning, support, competitor pricing)
- •Write explicit role instructions (PM expertise, strategy mindset)
- •Add behavioral constraints: candid feedback, pushback, less cheerleading
Day-to-day PM acceleration: building narratives, PRDs, and fast stakeholder answers
Amir explains his workflow loop: use the AI to ‘ping-pong’ toward an outline or narrative (PRD/product review), then re-upload the refined doc to strengthen the brain. The result is rapid, context-aware responses—useful for async stakeholder requests like marketing copy.
- •Use AI to iterate toward an outline/overview document
- •Re-upload finished artifacts to compound knowledge over time
- •Speed up ad-hoc asks (e.g., two-line descriptions for comms/marketing)
- •Benefit: quick, context-rich answers during busy Slack-driven days
Custom writing coach: importing Lenny/Wes Kao guidance to shorten Slack messages
After receiving feedback that his writing was too long, Amir builds a custom GPT to rewrite messages concisely while preserving a natural voice. He loads newsletter content and referenced books as style guidelines, then uses it daily by pasting drafts for rewrites.
- •Problem: strong writing but overly long updates and Slack messages
- •Solution: custom GPT with concise-writing guidelines from experts
- •Technique: copy/paste newsletters, save as docs, upload as knowledge
- •Daily habit: draft → paste into GPT → concise rewrite → better responses
AI for professional development: turning feedback into reusable coaching systems
Claire highlights a broader pattern: using AI to operationalize performance feedback and self-improvement. Amir notes that pre-AI, manager editing cycles were slow and momentum-killing; now, coaching is instant and repeatable, and shareable with colleagues.
- •Use AI to apply feedback continuously, not just at review time
- •Avoid slow manager back-and-forth editing loops
- •Create a repeatable coaching workflow that fits daily work
- •Share internal tools so others benefit from the same coaching system
Product interview prep with GPT voice mode: natural mock interviews + targeted improvement
Amir describes why voice mode is transformative for interview prep: it simulates real-time pressure and verbal delivery better than text. He uses either a custom interview-prep GPT or freestyle voice sessions, optionally adding his CV and the job description for context, and tracks skill improvements over repeated runs.
- •Voice mode > text for practicing real interview communication
- •Two modes: structured custom GPT or freestyle voice sessions
- •Add CV + job description for role-specific prompting and critique
- •Iterate: voice sessions identify weak spots (e.g., segmentation) and improve over time
Recap, ‘power user’ mindset, and lightning round: fixing AI when it’s wrong
Claire summarizes Amir’s workflows: data scraping + analysis, second-brain projects, writing coaching, and voice-based interview practice. In the lightning round, Amir argues the biggest miss for non-power-users is being unable to ‘be everywhere at once,’ and he shares his tactic for bad outputs: provide an example and a benchmark (sometimes with a raised voice).
- •Workflow recap: scrape → analyze → store in project brain → iterate documents
- •Power-user advantage: rapid answers and presence across many topics
- •When AI is wrong: paste output back and show an example of the desired result
- •Bonus: Amir is building a generalized scraper tool inspired by the show
