#004 - The 80/20 Rule Didn't Go Anywhere
November 25, 2025
Enjoying this newsletter?
Subscribe to get future issues
Issue #004: The 80/20 Rule Didn’t Go Anywhere
AI changed where the hard work lives, not whether it exists
November 25, 2025
First, Some Housekeeping
I missed an issue. Took a quick trip to Vietnam (I’m based in Thailand these days), and the newsletter slipped. Sorry about that.
Also: I’ve heard from a few of you that emails didn't land in your inboxes. The amazing Jesse from Bento has helped me resolve some deliverability issues, so hopefully that won’t happen again! If you missed the earlier issues, you can read them all at jkudish.com/newsletter. And if you want to make sure these land in your inbox, add my email ([email protected]) to your contacts or whitelist it.
One more thing before we dive in: I’d love to hear from you. Which topics have resonated most? What do you want me to write about? Hit reply and let me know. The feedback will really help me to know what resonates best and what to write about moving forward. 😄
The 80/20 Rule Didn’t Go Anywhere
You’ve probably heard some version of this: AI makes developers 10x faster. Ship in hours what used to take weeks. The future is here.
There’s truth in it - writing code has certainly sped up. But the reality is more nuanced.
After months of building with AI daily, I’ve noticed the 80/20 rule hasn’t disappeared. It’s just shifted where it applies.
Three Patterns
1. The Role Flip
My time distribution inverted. I used to spend maybe 80% of my time writing code and 20% planning and reviewing. Now it’s the opposite. I spend most of my time on specs, prompts, and reviewing what AI produces. The actual “building” is a fraction of the work.
This isn’t a complaint. At all. I actually love the architecting phase. It’s one of my favourite phases of any project and probably where I thrive the most.
2. The Last Mile Problem
AI gets me 80% of the way there. Solid foundation, working code, reasonable structure, relatively bug free (it has its bad days though!). But that final 20%? Still mine.
Edge cases. Integration quirks. The subtle UX decisions that only surface when you’re actually using the thing. A detailed spec helps, but it can’t capture everything no matter how detailed you are. Building software requires using the software and seeing how the interactions flow - that feedback loop is crucial.
3. The Effort Illusion
This one’s sneaky. I’ll look at a feature and think “we’re 80% done.” But in terms of time remaining? I’m actually only 20% through. The first 80% happened so fast that I underestimated what was left.
AI compresses the easy part. Which means you hit the hard part sooner - and it can sometimes feel like it takes forever because you expected to be done already.
The Meta-Insight
Here’s what I think is actually happening:
AI didn’t eliminate the 80/20 rule. It revealed where the hard work actually lives.
The stuff we used to spend 80% of our time on - the boilerplate, the scaffolding, the “just typing it out” work - that was never the real work. It was overhead. AI squished it way down, and suddenly we’re face-to-face with the 20% that always mattered: judgment calls, edge cases, integration, polish, deciding what to build in the first place.
We weren’t doing more valuable work before. We were just busy.
Knowing what to build, when to build it, and how to approach it - that’s still where experience matters. AI can’t tell you which feature to prioritize or whether your architecture will hold up in six months. That judgment is yours.
A Real Example
I’ve been building number porting for Tether - the feature that lets users bring their existing phone number into the app. Wrote a detailed spec. Handed it to Claude. A few hours later, I had working code.
Then I actually tried using it.
The happy path? Fine. But what happens when someone enters their number in a weird format? The loading state while we check with the carrier felt too abrupt. The confirmation message was accurate but read like a robot wrote it. Error handling existed but didn’t actually help anyone fix the problem.
And some stuff I just hadn’t thought about at all. Turns out number porting requires a signed document. I knew that abstractly, but I hadn’t figured out how to actually handle it in the app until I was staring at the flow. That’s not a spec failure - it’s the kind of thing that only crystallizes when you’re in the middle of building.
None of this was wrong, exactly. Claude followed the spec. But the feature wasn’t done.
That last 20% - tweaking copy, smoothing out micro-interactions, handling the edge cases you only discover by clicking around, making decisions that couldn’t be made upfront - ended up taking longer than the initial build.
The Takeaway
AI doesn’t make you 10x faster at everything. It makes you 10x faster at the parts that weren’t the bottleneck.
The hard problems are still hard. You just get to them sooner.
Plan for it. Budget time for the 20% that remains stubbornly human. And don’t mistake a working prototype for a finished product.
What I’m Building
Tether is an SMS gateway for digital nomads and teams who need reliable access to verification codes and other messages. You get a phone number, and texts show up in Telegram - perfect for 2FA codes when you’re abroad or when your team needs to share access to verification messages. Soon: Email, Slack, Discord, and full team sharing.
Right now I’m in that last 20% phase that will probably take 80% of the time - polishing flows, handling edge cases, architecting for teams. I’m having a lot of fun perfecting it. Launch is coming soon.
If you want early access, sign up at tethermobile.com.
I Have Bandwidth
Quick note: I’m taking on one new client project for December/January.
I do Laravel development and fractional CTO work. I’ve spent 18+ years building products - everything from e-commerce platforms to AI automation pipelines. I’m remote, async-friendly, I ship quickly and focus on how I can deliver value in any project I’m involved in.
If you’ve got something interesting, reply and tell me how I can help!
Cool Stuff From the Internet
difit - Code Review for the AI Era
A local code review tool that shows git diffs in a GitHub-style interface. The interesting part: it’s designed for AI workflows. You can add comments to the diff and copy them as prompts.
I’m working on some developer tools myself (more on this soon, I promise!), and this is the kind of thing I love seeing. AI isn’t just changing how we write code - it’s creating demand for new specialized tools. A renaissance for developer craftsmanship. I’m so excited for it. It’s like Christmas several times a year for us nerds, with so many new tools.
Build Your Own Database
An interactive tutorial that walks you through building a key-value database from scratch. You end up with an LSM tree - the same data structure behind LevelDB and DynamoDB.
What makes it great: the visualizations. You click buttons, add records, and watch the underlying structure change in real time. A truly beautiful example of teaching through doing.
Want to feature something cool you saw or that you’re working on? Hit reply or mention me on X (@jkudish).
Until Next Time
I’d genuinely love to hear how this lands for you. Have you felt the 80/20 shift? Are you spending more time planning and reviewing than you used to? Hit reply - I read everything.
And if you’re in the US or celebrate - happy Thanksgiving! Hope you get some time to unplug and recharge.
Keep shipping, Joey
P.S. If you know someone who might enjoy this, send it their way. They can subscribe at jkudish.com/newsletter.