I like building things, solving real problems, and turning ideas into something useful.
Afghanistan
Joined March 2022
- Tweets679
- Following311
- Followers183
- Likes1.3K
Claude is getting Whiteboards, Sheets and Data.
I think the interesting shift here is that AI products are moving beyond answering questions and generating text.
They're trying to become places where actual work happens.
The challenge now is making these tools genuinely useful without making the workflow more complicated.
5
Vercel's new Schedules feature is interesting.
Applications can create and change scheduled jobs at runtime without a redeploy.
For AI agents, that's a pretty important building block.
Understanding "do this tomorrow" is one thing. Having reliable infrastructure to actually do it tomorrow is another.
1
29
I miss those days where cursor usage and grok bot was separate.
Grok itself is a token burner, let alone if you use grok 4.7 to dump your tokens into its reasoning.
2
1
39
After grok downnshifted 2 times I don't have much of a high hope. Not gonna lie I do want it to have that grok 4.5 moment again.
Elon Musk just revealed a massive update on Grok 4.8, and SpaceXSI has completely rebuilt its reinforcement learning and inference infrastructure from scratch
The engineering team has abandoned JAX and SGLang, rewriting the entire RL and inference stack in C/C++
Here’s where Grok 4.8 stands:
• Pre-training is complete
• Major cleanup of supervised fine-tuning data
• Reinforcement learning now underway
• Evaluation scores are starting to climb
• Soft rollout to Grok Bot users planned for later this week
And Elon also gave another imp update about Grok 4.8:
“Grok 4.8 will bear no resemblance to 4.7 (thankfully)”
He also explained why the development took longer than expected:
“It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special”
This is a fundamental engineering overhaul, not just another model upgrade
SpaceXSI has rewritten the underlying infrastructure to build a much stronger foundation for Grok’s future development.
And the first Grok Bot users are expected to get their hands on it this week
13
Good architecture isn't about preparing for every possible future requirement.
It's about making the changes you're likely to need without turning the whole codebase upside down.
I've never liked adding complexity just because something might become a problem later.
6
Codex CLI now has an agent command center for managing sessions and tracking unread results.
I like the direction here.
Once you're running multiple coding agents, the hard part becomes keeping track of what each one is doing and what still needs your attention.
That workflow matters just as much as the models.
5
5
119
Anthropic just dropped Claude Fable 5.2.
Apparently, they used it to migrate a production service from Python to Rust, and startup is now 26x faster.
Discord is moving to Rust too.
At this point, I'm convinced someone's rewriting their coffee machine in Rust.
44
GitHub tested 5,000 AI agents working on a single repository, with 7,200 pushes per second.
What caught my attention wasn't the number.
They've redesigned Git storage so most writes can happen independently, while reference updates handle the coordination.
AI is starting to change the infrastructure behind software development, not just how we write code.
11
X's engineering team published an interesting breakdown of their web app rebuild.
What caught my attention wasn't just the performance improvement.
It was how much of that improvement came from changing the architecture and when content gets delivered.
A good reminder that frontend performance isn't always about optimizing JavaScript.
13
A product can have a lot of features and still feel incomplete.
Sometimes the problem isn't what's missing.
It's that the existing features don't work together as smoothly as they should.
15
One thing that annoys me in software is when an action fails and the app gives you absolutely no useful information.
"Something went wrong."
Okay, but what?
Did my session expire? Do I lack permission? Is the server unavailable?
Good error messages aren't a small detail. They're part of the product.
1
18
Database migrations deserve more attention in code reviews.
A bad function can usually be fixed with another commit.
A bad migration can leave you dealing with broken data, failed deployments, and a rollback that isn't nearly as simple as you expected.
I'd rather spend an extra hour reviewing the migration.
1
27
There's a big difference between:
"Fix the authentication bug."
And:
"This endpoint returns 403 for an authenticated user. Trace the authorization check, explain why it fails, and identify the smallest safe fix before editing."
Same AI model. Very different instructions.
The second gives it boundaries instead of just a goal.
27
I don't mind AI agents making large changes when there's a good reason.
What bothers me is a small bug fix turning into a rewrite of unrelated files.
Sometimes the smaller patch is the more thoughtful solution.
How are you reviewing the scope of agent-generated changes?
28
Before merging an AI generated refactor, check four things:
- What behavior actually changed?
- Did it touch files outside the original scope?
- Which tests cover the change?
- Could the same result be achieved with fewer edits?
A passing test suite is important. It doesn't automatically mean the refactor was a good idea.
1
2
45
Been using @t3dotcodes for over a month and I can't stop shipping the experience makes the flow way enjoyable.
@theo One of my favorite updates is the context compaction which is really nice to have when you work on a 500k line codebase, and AI like to drift.
2
67
I maintain a codebase with over 500k lines.
One thing I've learned is that giving an AI agent more code doesn't necessarily give it better context.
A failing test, the relevant files, and a clear description of what should happen are usually far more useful than dumping half the repository into a prompt.
1
2
87
We had two reasonable ways to structure a feature.
Here's the tradeoff that mattered more than the framework choice.
With AI framework is solved.
33