Abhi Basu

Product Leader · AI Builder

I use AI to do my job better. This site documents how — from personally building 76 agent skills that automate my daily workflow, to writing about what actually works when you pair a product manager with an AI copilot.

🧭 AI PM OS ✍️ Thoughts 🛠 Projects ↗ GitHub
Abhi Basu

Abhi Basu

Staff Product Manager @ Square
📍 San Francisco Bay Area
🔗 LinkedIn 🔗 GitHub 📄 Download CV ✉️ contact@abhiroopbasu.com

I'm a product leader with 12+ years building software across B2B SaaS, fintech, and payments. I started my career as a lawyer in Singapore, pivoted into marketing at a 30-person startup, and worked my way up through product marketing and product management at companies from early-stage to enterprise.

Today I lead checkout and payments at Square — the flows that run every time a seller takes a payment on a Square terminal. I also build AI tools that make me better at my job. I've personally created 76 AI agent skills that automate everything from morning triage to writing PRDs to analyzing data, and I built amp-mem, a persistent memory system for AI agents.

12+
Years in Tech
XXM
Monthly Checkout Volume ↑
50%
App Download Growth
76
AI Skills Created
Square Staff Product Manager 2022 → present
  • Own checkout and payments across Android & iOS — increased merchant checkout success rate by 1%, driving multi-million monthly volume increase
  • Shipped generative AI onboarding (GPT-4) that increased conversion by 6% — replaced 20+ static questions with a single free-form text box
  • Led cross-platform Neighborhoods initiative spanning Square and Cash App
  • Grew Dashboard app downloads 50% via growth campaigns; built mobile sales and employee reporting
Zendesk Director of PMM → Director of Product 2014 → 2022
  • Managed 5 PMs and 7 PMMs across growth and omnichannel products
  • Increased retention by 1% and revenue by >$2M/year; launched first self-service onboarding for 100K+ customers
  • Grew net revenue of Suite product 15% through PSM-driven pricing strategy; drove >$2.5M pipeline from targeted campaigns
  • Led enterprise go-to-market post-acquisition; increased live chat ASP by 30%
Zopim Product Marketing Lead 2013 → 2014
  • First marketing hire at a 30-person startup (acquired by Zendesk in 2014)
  • Grew organic traffic 135% through content and SEM strategy
Potato Productions Digital Strategist 2011 → 2013
  • Digital strategy and social media consulting for brands across Southeast Asia
Amica Law Lawyer 2010 → 2011
  • Admitted as Advocate of the Supreme Court of Singapore
  • Intellectual property law — patents, trademarks, copyright; advised startups on licensing
🎓 Education: LLB (Upper Second Class Hons) from the University of Warwick, UK. International Baccalaureate from Mahindra United World College of India.

🌍 Background: Born in India, practiced law in Singapore, built products across Asia and the US. Fluent in English, conversational in Bengali and Hindi. PADI rescue diver, chess society president, and former TEDx organizer.

Featured Project: AI PM OS

An operating system for product work, built around a Chief of Staff agent, durable workstreams, and resumable parallel sessions.

Landing page

See the full system

A walkthrough of the operating model: start-of-day planning, CONTEXT.md continuity, lightweight derived state, and the public starter repo.

overview
GitHub repo

Clone the public export

A sanitized, tool-agnostic version of the system with starter workstreams, repo-local skills, and a simpler today-plan.json-driven state model.

open source
Three-part series

Read the story behind it

Why I built it, how the launcher works, and why durable context plus orchestration feels different from a normal chatbot workflow.

thoughts
Explore the AI PM OS → View repo ↗

Thoughts

Notes on building with AI — what works, what doesn't, and what I've learned

2026-04-12

Why the AI PM OS Feels More Powerful Than a Chatbot

The leap comes from orchestration, durable context, and startup routines around the model, not just the model itself.

ai-pm-os
2026-04-12

How the AI PM OS Spins Up My Entire Workday

A technical walkthrough of the launcher, the Chief of Staff planner, the public state layer, and resumable cmux workstreams.

ai-pm-os
2026-04-12

I Built an AI PM OS

Why I turned a collection of AI skills into a product-management operating system with durable workstream context.

ai-pm-os
2026-03-24

We Built an AI Skill That Writes Product Requirements Docs

How a team went from markdown-first specs to an agent skill that writes PRDs with auto-referenced internal sources — in two weeks.

product-management
2026-03-24

How I Automated Every Workflow with AI Skills

A practical guide to building 35+ AI agent skills that automate morning triage, email, project management, and more — no engineering degree required.

workflow
2026-03-23

Cutting 80% of the Noise from My AI Memory System

After distilling 2,718 observations, found ~90% were noise. Seven plugin changes cut volume by ~80% while preserving signal.

memory-system
2026-04-06

I Stopped Telling My Agent Which Tool to Use

How the auto-pilot routing layer turned a long menu of skills into one assistant that can pick and chain the right tools automatically.

orchestration
2026-04-06

The File That Runs My Entire AI Setup

The single AGENTS.md file that defines identity, boundaries, startup behavior, routing defaults, and how the whole AI environment behaves.

agents-md
View all posts →

What I've Built

Original skills I created for my daily PM workflow — 76 total on GitHub

☀️
start-of-day
Morning triage: Slack → Gmail → Calendar. Surfaces unreads, drafts replies, shows schedule.
"start my day"
✅
todo
Persistent to-do list with proactive reminders. Auto-captures from Slack, Gmail, Calendar, Linear.
"add a to-do to review the spec"
📋
plan-to-linear
Convert a structured plan into Linear issues and projects in bulk.
"convert this plan into Linear tickets"
📥
feature-request-scanner
Daily scan of Slack channels for feature requests. Auto-classifies, scores, deduplicates.
"scan for new feature requests"
📝
writing-requirements-docs
Write PRDs from rough notes with auto-sourced evidence.
"write a PRD from these notes"
📈
data-analyst
Full analysis pipeline with follow-up questions, charts, and visualizations.
"analyze abandonment trends"
🚀
pr-manager
Commit changes, create/update PRs. Supports Graphite stacked PRs.
"commit and create a PR"
🧠
memory
Persistent cross-session memory — captures observations, generates summaries.
"search memory for auth decisions"
🔬
rpi
Research → Plan → Implement for complex, multi-step tasks.
"rpi: refactor the auth module"
📅
reviewing-calendar
Visual weekly calendar with conflict detection and color coding.
"review my calendar this week"
📢
shipped-announcements
Craft and post launch announcements to Slack using Block Kit.
"write a shipped post for the new feature"
📬
cash-rounding-responder
Triage and draft replies to feature feedback emails. Classifies, prioritizes, drafts responses.
"respond to feature feedback emails"
📄
drafting-docs
Draft and format Google Docs with consistent styling.
"draft a strategy doc"
📑
slack-saved-triage
Prioritize and group Slack saved messages. Triage your save-for-later queue.
"triage my saved messages"
👥
manager-slack-summary
Weekly summary of Slack conversations with your manager. Groups by topic.
"summarize my slack with my manager"
▶️
linear-to-execution
Pick up a Linear issue and prepare for implementation.
"work on PROJ-42"
📊
project-status
Aggregate project status from Slack, Drive, GitHub, and feature flags.
"what's the project status?"
🕰️
historical-info
Search across sources to find what you've been working on.
"what have I worked on this week?"
🗺️
blueprint-intake
Create roadmap projects from Slack threads. Digests discussion, populates all fields.
"add this to the roadmap"
🔍
blueprint-project-status
Audit and update project statuses in bulk. Cross-reference connected resources.
"audit project statuses"
📝
blueprint-project-update
Draft and add weekly project updates from connected resources.
"write a project update"
🔄
blueprint-status-update
Sync project statuses from Linear, Slack, and Jira.
"sync project statuses"
📊
weekly-status-summary
Generate consolidated weekly summary from project updates and post to Slack.
"post weekly status summary"
👀
monitoring-prs
Watch open PRs for new comments and CI failures. Auto-fix and respond.
"monitor my PRs"
💬
address-pr-comments
Fix unresolved PR comments, commit per-comment, resolve threads.
"address PR comments"
✅
code-review-general
Address code review feedback — prioritize human comments, verify changes.
"fix the review feedback"
🗺️
walkthrough
Explore and visualize codebase architecture with diagrams.
"walk me through the payment flow"
🔧
check-ci
Check, debug, and fix CI builds and test failures. Loops until green.
"fix the CI failure"
📦
cloning-squareup-repos
Clone repositories using the correct SSH remote and conventions.
"clone the repo"
📦
hermit
Manage hermetic developer tool packages per-repository.
"install the tool via hermit"
🔗
protos
Search and inspect protocol buffer definitions, messages, services, and enums.
"find the proto definition"
📐
spec-creator
Create product specs using structured templates, synced to Linear.
"create spec for new feature"
🔎
feedback-searcher
Search and synthesize user feedback across multiple sources.
"what are users saying about onboarding?"
🧩
prototype-builder
Scaffold and build interactive HTML prototypes for product features.
"build prototype for the new flow"
🚀
launch-a-product
Guide PMs through GTM launch readiness. Validates fields, crafts positioning, drafts comms.
"check launch readiness for the new feature"
✅
test-plan-creator
Generate test plans and acceptance criteria from product specs.
"create test plan for the redesign"
📈
data-analyst
Full analysis pipeline with follow-up questions, charts, and visualizations.
"analyze abandonment trends"
🏪
ecom-great-stores
Search and surface example seller sites by industry, feature, or use case.
"show me bakery site examples"
🔬
ecom-research
Search seller research insights: quotes, site reviews, competitive analysis, CSAT.
"what do sellers say about ordering?"
🛡️
trust-feature-validation
Validate feature behavior during rollouts. Spot-check users, audit data.
"validate the feature rollout"
🏗️
staging-account-builder
Provision full staging accounts with payments, catalog, orders, and labor data.
"create a test restaurant account"
🤖
auto-pilot
Routes any request to the right skill(s) automatically.
(always active)
🐝
swarm
Multi-perspective investigation with adversarial challenge and consensus.
"investigate the latency spike"
♻️
ralph-loop
Iterative work-review loop using two AI models.
(complex revision tasks)
🧬
kb-distill
Distill raw observations into structured knowledge notes.
"distill my recent work"
💬
jack-guidance
Concise, direct coaching for internal communication. High-signal, low-fluff.
"jack guidance on this status update"
🤖
eng-ai-chat
Search internal knowledge grounded in app context. AI-powered answers.
"how does the auth service work?"
☁️
blox
Delegate work to cloud workstations. Launch, provision, and manage remote dev environments.
"run this on a cloud workstation"
📚
dev-guides
Search and browse internal developer documentation and guides.
"find the setup guide"
🔗
go-link
Resolve internal go/ shortlinks to full URLs.
"what does go/oncall point to?"
🎫
jira
Manage Jira issues, transitions, sprints, and workflows.
"show my open Jira tickets"
View all 76 skills →

amp-mem — Persistent Memory for AI Agents

Most AI agents forget everything between sessions. I built a memory system that doesn't. amp-mem is a ~500-line bash CLI backed by SQLite + FTS5 that gives AI agents persistent, searchable memory — no vector database, no cloud service, no embeddings. The LLM itself is the semantic engine.

16
CLI Commands
FTS5
Search Engine
0
External Deps

🔍 Passive Capture

Auto-records decisions, discoveries, and workflows from tools like Slack, Gmail, and Linear as you work

🧬 Knowledge Distillation

Compacts raw observations into structured notes — surfaces patterns across hundreds of sessions

⚡ Context Injection

Automatically primes new sessions with relevant past context — the agent remembers what matters

📦 Zero Infrastructure

Single SQLite file. No vector DB, no API keys, no cloud. Works offline, backs up with git

Deep dive → View source → 
$ amp-mem search "checkout flow"

── Results (3 of 47 observations) ──

📌 decision 2026-03-22
  Checkout skip-screen removed after
  A/B test showed 12% lift in completion

📌 discovery 2026-03-18
  Cash rounding applies AFTER tax calc,
  not before — confirmed in register-svc

📌 workflow 2026-03-14
  New flow: PR → playpen deploy → smoke
  test via agent-browser → merge

$ amp-mem stats

  Observations:  2,718
  Sessions:      342
  Compacted:     89 notes
  DB size:       4.2 MB

$ amp-mem compact
  Distilling 51 new observations...
  ✓ 7 knowledge notes created

Under the Hood

The systems that make it all work

🧭

AI PM OS

The full product-work operating system: Chief of Staff planning, durable workstreams, resumable CMUX sessions, and a public starter repo you can adapt.

Explore the system →
🧠

Persistent Memory

SQLite + FTS5 memory system that persists across sessions. Auto-captures decisions, discoveries, and workflows. No vector DB — the LLM is the semantic engine.

Learn more →
⚡

Amp Setup

The full guide to getting started — installing Amp, adding skills, configuring AGENTS.md, and setting up cross-session memory.

Setup guide →
🤖

AGENTS.md

The real configuration file powering this setup — identity, session startup, memory, to-do system, tool preferences, and output style. Sanitized for public sharing.

View the full file →