Most important part missing. Switch to @useblacksmith 🫡
Agents were shipping code faster than our CI pipeline could keep up.
Our team optimized our pipeline, leading to a roughly 15% faster test suite and 50% reduction in runner-time spent per test, all while our test suite grew 4x in size.
@moofeez explains how:
linear.app/now/ci-bottleneck…
1
2
7
952
Now that some features have been added (and things like cmd + f work), I can finally play with some ideas I've had around making PR reviews more effective - yes, i still read code.
A few experiments 🧵
Was thinking about a few enhancements to codiff. But @t3dotcodes had some of them already. So I downloaded t3code today, and brought my favourite part of codiff - walkthrough into t3.
Not loving the idea of t3code yet, but maybe it will grow on me with this.
2
127
5. Funny enough, the more I look for ways to understand a change beyond just reading the diff - given I don’t fully trust an AI agent to do the review for me yet - the more I keep finding ways to give the agent better context.
And at some point, that might actually make it better at reviewing the PR than I am. 😅
1
17
And another s/o to @t3dotcodes for being open source (and easy to fork locally).
Every one of these experiments is basically a prompt away, and the product is already good enough that I can use it in my actual day-to-day work while trying these ideas as a POC.
That makes it a pretty fun place to test out what a better PR review workflow could look like.
16
Was thinking about a few enhancements to codiff. But @t3dotcodes had some of them already. So I downloaded t3code today, and brought my favourite part of codiff - walkthrough into t3.
Not loving the idea of t3code yet, but maybe it will grow on me with this.
1
1
2
253
Anyone make the switch from graphite stacks to @github stacks? Smooth transition? Be nicer to stay on a single platform. Code reviews are better in graphite too, but devs using alternative tools anyways (codiff etc) be a nice saas line item to drop
30
Code quality used to be enforced by self-interest. The person adding the boolean flag was going to be the person maintaining it. The agent never has to live in the codebase it leaves behind.
Where I land on PR reviews now that code is cheap:
"Just merge it, the AI will fix it." It works. It always works in the beginning. But i still think there is purpose to reading code.
1
87
"Just merge it, the AI will fix it." It works. It always works in the beginning. But i still think there is purpose to reading code.
9
2
6
1,481
13 years ago, Apple asked a question that hits me harder today than ever:
If everyone is busy making everything, how can anyone perfect anything?
Every day, I see a world flooded with effortless AI slops. I’m realizing that to build something truly great today, you need a level of focus the world has never demanded before.
It feels like trying to write a quiet, beautiful song in a room full of loud noise.
105
472
3,944
139,004
I like codiff. But I think I rather have it as an extension (or built in) my IDE. Sometimes I just want to ctrl click into a function while reviewing, or 1 line changes I might as well do myself instead of comment and pass to AI.
1
98