The 10 Days That Saved Me 3 Weeks for Chrome Extension
- Sangamesh Gella
- Nov 18
- 4 min read
I shipped my Chrome extension in 3 weeks. It failed at authentication. I rebuilt it in 24 days. It worked perfectly. The difference? I spent 10 days not writing code.
This is the story of how I learned paradoxical productivity the hard way... and why it changed everything about how I build with AI.
The First Ship: Fast Equals Broken
Three weeks of coding with Claude Code. I jumped straight into implementation. Build the detector. Build the UI. Build the features. Ship it.
No spec. No system. Just momentum and determination.
I was obsessed with the outcome. Ship fast. Get it live. Iterate later.
The approach felt productive. Lines of code every day. Progress bars are moving. Features shipping. Then I published it. Authentication failed when users tried to access the extension from the same Salesforce page.
Not sometimes. Every time.
I stared at the error logs. I had no idea how Salesforce session authentication actually worked across domains. I'd spent three weeks coding... but I didn't understand the system I was building for.
That's when I realised what I'd violated: Outcome vs. Process. I was goal-obsessed with no system underneath.
I was also violating Flexibility. Locked into one approach: code-first, force it to work, grind through blockers.
And I was ignoring Life-Work Balance. Three weeks of hustle with no space to think or breathe.
The cost? A broken extension and a burnt-out builder.
The Step Back: Doing Nothing Was Doing Everything
I had a choice.
Rebuild it the same way. Code faster. Try harder.
Or... stop.
I did three things that felt completely wrong:
I stopped coding
I admitted I didn't know enough
I gave myself space to learn
Ten days of pure research. Zero lines of code. This was Mindset Shift in action. What if slower is actually faster? What if I don't know enough yet?
This was an Unconventional Strategy. The hustle culture playbook says: code harder, ship faster, figure it out as you go. I did the opposite.
And this was Life-Work Balance. Instead of grinding through weekends and all-nighters, I gave myself permission to step back.
I dug into Salesforce authentication architecture.
The discovery: session authentication requires a specific chain.
New tab → host domain → convert to old domain → convert to enhanced domain → extract session ID → use SID for auth.
Each domain has different authentication rules. You can't just grab a session ID from the current page and make it work.
I mapped it out. I documented it. I understood the system. It felt slow. Unproductive. Like I was wasting time. But I was finally moving in the right direction.
The Rebuild of my Chrome Extension: Clarity Unlocked Speed
Version 2 wasn't built on hope. It was built on a system. I applied three more FORMULA principles:
Outcome vs. Process (again, but correctly this time): I built a process first. Linear tickets became my system for breaking down work.
Routines & Rituals: I created a daily workflow ritual:
Morning: review tickets
Afternoon: write specs for new features
Evenings & Weekends: let Claude Code execute the clear tasks
Flexibility: When a ticket didn't work, I adjusted the spec. I didn't force the code.
Ten feature tickets. Each had 2-3 tasks. Each task had clear acceptance criteria.
The auth flow looked completely different:
Version 1 approach:
"Build auth" → vague conversation with Claude → code → debug chaos → repeat
Version 2 approach:
Ticket 1: Research Salesforce domain conversion rules (spec complete)
Ticket 2: Implement SID extraction logic (build with context)
Ticket 3: Test cross-domain authentication flow (verify it works)
I connected Linear to Claude Code through MCP. The tickets became specifications. The specifications became implementation.
Here's the time breakdown that tells the whole story:
Version 1: 21 days of coding → 0 working extensions
Version 2: 10 days of research + 14 days of structured building → 1 working extension
The paradox was fundamental. The 10 days I spent "doing nothing" made the next 14 days actually work.
The Principle: AI Forces Clarity
This wasn't just about a Chrome extension. It was about how AI development actually works. AI agents don't eliminate thinking. They reveal when you haven't done enough of it.
Claude Code can build anything... if you tell it what to make clearly enough.
The paradox: AI tools promise speed, but they require upfront clarity to deliver it. You can't skip the thinking. You can only choose when to do it.
Version 1: I tried to think while coding. It was chaos.
Version 2: I thought first, then coded. It was clean.
This is Adaptation & Action in practice. One failure taught me a system I now use for everything:
Salesforce automation: spec the flow before building it
Agentforce sub-agents: map the orchestration before coding
Any agentic workflow: specification-first beats code-first
The shift from code-first to spec-first changed my entire development approach.
What I Learned About Paradoxical Productivity
This story maps to every principle in the FORMULA I am building this year:
F (Flexibility): I had to break my code-first habit and completely pivot my approach
O (Outcome vs. Process): Building a system delivered what forcing an outcome never could
R (Routines & Rituals): Linear + MCP + Claude Code became my repeatable workflow
M (Mindset Shift): Slower felt wrong, but turned out to be the only thing that worked
U (Unconventional Strategy): Ten days of "doing nothing" was doing everything that mattered
L (Life-Work Balance): Stepping back instead of grinding saved both the project and my sanity
A (Adaptation & Action): One failure taught me a framework I use every day now
The 10 days I spent not coding weren't wasted time. They were the investment that made the next 14 days—and every build after—actually work. Paradoxical productivity isn't a theory. It's what happens when you stop forcing speed and start building clarity.
Sometimes the fastest way forward is slowing down long enough to see the path.
If you liked reading this, I am sure you will love my newsletter, where I share deep dives on topics like Salesforce + Dev life, Tech & AI trends, Productivity experiments, Prompting techniques, New AI tools I test, and What's on my mind in tech, and it's free, no spam!
All you have to do is subscribe to get the good stuff in your inbox.
Comments