Compounding Decisions for the Age of AI
Subscribe
Sign in
Home
Archive
About
Latest
Top
How my agent splits work across my Claude and ChatGPT subscriptions
I built pmmate, a small supervisor agent that routes my PM work to Claude Code or Codex based on a few rules such as task scope/ complexity, how much…
Oct 10
•
Sriram Natarajan
1
1
Why coding agents make segment choice more important
I frequently see a similar question in some of my PM Slack groups: if AI reduces the cost of building software, do companies still need to pick or focus…
Oct 4
•
Sriram Natarajan
1
1
September 2026
What Amazon blocking Muse tells us about agentic commerce
Last week, I wrote that I might eventually let Muse buy a movie or concert ticket for me.
Sep 27
•
Sriram Natarajan
1
1
Why we will use different AI agents for different parts of our life
Consumer AI products differ across two dimensions: how much the user must define and how much authority the agent has.
Sep 21
•
Sriram Natarajan
Painful lessons from upgrading to OpenClaw 2026.09.01
The upgrade appeared to work. Then my agents slowed down, approval cards disappeared, and I spent 48 hours trying to recover the system.
Sep 21
•
Sriram Natarajan
1
1
My first real test of Grok Bot after six months with OpenClaw
I asked Grok Bot to summarize my LinkedIn and X feeds, then compared the experience with the agent harness I use every day.
Sep 14
•
Sriram Natarajan
1
How I got my AI agent to stop changing its mind
My investing agent kept changing its recommendation on the same numbers. Surprisingly, the fix wasn't a smarter prompt - and it works for any agent.
Sep 7
•
Sriram Natarajan
2
August 2026
How I used my agent to clean up my Gmail
I was close to hitting the 15 GB size limit in my Gmail.
Aug 30
•
Sriram Natarajan
1
How I set up my agent to be my accountability partner
AI Agent can help accomplish your goals and can also hold you to be accountable to your goals. Here is how I set up my agent to be my accountability…
Aug 23
•
Sriram Natarajan
2
1
I gave my OpenClaw a gate, and it started refusing me
The fix for a too-eager AI agent wasn't a better model. It was teaching it when to stop and ask me.
Aug 17
•
Sriram Natarajan
1
1
How I use AI to find my own growth opportunities
I pointed Claude Cowork at 18 months of my Slack and email and asked it to rank me against Amazon's Top 10 leadership principles.*
Aug 9
•
Sriram Natarajan
1
1
When AI makes building cheap, the job is knowing what's worth building
A customer discovery call on Monday. A working demo on Wednesday. What changed in between wasn't my speed.
Aug 2
•
Sriram Natarajan
2
1
This site requires JavaScript to run correctly. Please
turn on JavaScript
or unblock scripts