11,000 pull requests in one month.
Revert rate around 0.5%.
Most AI coding is pure slop.
Lauren Tan’s team at SpaceXSI just showed how they turned it into something that actually ships clean.
Free talk. From the people running Grok Bot agents at scale.
Save it. Watch tonight.
Changes how you think about agents writing code.
You’ve been letting agents dump code and hoping it holds up. This is the opposite approach.
Follow @codewithimanshu for more high-signal content that actually moves your skills forward.
↓
Lauren Tan (poteto).
She talks about the trust pyramid and codebase quality first.
Not “just prompt better.”
They built the foundations so agents can move fast without everything falling apart.
Architecture that enforces itself. Lint rules instead of comments. Tests that actually catch stuff.
Follow @codewithimanshu , I break down talks like this.
↓
Turning slop into gold.
The whole point of the talk.
Most people scale agents and quality dies. More PRs, more bugs, more reverts, more pain.
They did 11k PRs and barely anything got reverted.
That’s not luck. That’s systems.
Runtime verification. Feature maps. Formal verification with TLA+ and Lean for the hard parts.
They spend tokens on checking the work, not just generating more of it.
↓
Cursor agents + pstack.
She shows how they turn the boring engineering foundations into actual software factories.
Agents don’t just write code. They work inside rules that make bad code hard to land.
Dune for enforceable architecture. Replacing “please don’t do this” comments with lint rules that block it.
This is the part most teams skip. They jump straight to agents and wonder why the codebase turns into mess.
More breakdowns of this kind of stuff on @codewithimanshu.
↓
Foundations before software factories.
She keeps coming back to this.
You can’t just throw agents at a messy codebase and expect magic.
Build the constraints first. Make the architecture enforce itself. Verify at runtime and formally where it matters.
Then the agents can go fast and the revert rate stays near zero.
11,000 PRs. 0.5% reverts. That’s the result of doing the boring work first.
↓
Timestamps if you want to jump:
0:00 - Turning slop into gold
2:03 - The trust pyramid and codebase quality
5:00 - Dune and enforceable architecture
6:56 - Replacing comments with lint rules
9:40 - Runtime verification and feature maps
10:50 - Formal verification and UI testing
11:51 - Spending tokens on verification
13:32 - 11,000 PRs and a 0.5% revert rate
15:11 - Foundations before software factories
Watch the full thing.
Then look at your own agent workflow and ask where the real constraints are missing.
Follow @codewithimanshu for more high-signal content that actually moves your career forward.
Auto-apply to AI-matched real-time jobs with a tailored resume in 2 clicks.
→ https://t.co/vDQEY9M5nv
→ Shows jobs from across the internet in 1 place
.
Join My Community→ https://t.co/s8NPsoA69F
Saying goodbye to the Los Santos trio at sunset before hopping in the pickup truck with the new Vice City duo actually hurts 😭🌅
Btw, CapCut is doing 90% off your first month for Black Friday right now if you wanna make your own GTA 6 edits 👇
https://t.co/ESjVv24XWc
#GTA6 #CapCutDidThat
11,000 pull requests in one month.
Revert rate around 0.5%.
Most AI coding is pure slop.
Lauren Tan’s team at SpaceXSI just showed how they turned it into something that actually ships clean.
Free talk. From the people running Grok Bot agents at scale.
Save it. Watch tonight.
Changes how you think about agents writing code.
You’ve been letting agents dump code and hoping it holds up. This is the opposite approach.
Follow @codewithimanshu for more high-signal content that actually moves your skills forward.
↓
Lauren Tan (poteto).
She talks about the trust pyramid and codebase quality first.
Not “just prompt better.”
They built the foundations so agents can move fast without everything falling apart.
Architecture that enforces itself. Lint rules instead of comments. Tests that actually catch stuff.
Follow @codewithimanshu , I break down talks like this.
↓
Turning slop into gold.
The whole point of the talk.
Most people scale agents and quality dies. More PRs, more bugs, more reverts, more pain.
They did 11k PRs and barely anything got reverted.
That’s not luck. That’s systems.
Runtime verification. Feature maps. Formal verification with TLA+ and Lean for the hard parts.
They spend tokens on checking the work, not just generating more of it.
↓
Cursor agents + pstack.
She shows how they turn the boring engineering foundations into actual software factories.
Agents don’t just write code. They work inside rules that make bad code hard to land.
Dune for enforceable architecture. Replacing “please don’t do this” comments with lint rules that block it.
This is the part most teams skip. They jump straight to agents and wonder why the codebase turns into mess.
More breakdowns of this kind of stuff on @codewithimanshu.
↓
Foundations before software factories.
She keeps coming back to this.
You can’t just throw agents at a messy codebase and expect magic.
Build the constraints first. Make the architecture enforce itself. Verify at runtime and formally where it matters.
Then the agents can go fast and the revert rate stays near zero.
11,000 PRs. 0.5% reverts. That’s the result of doing the boring work first.
↓
Timestamps if you want to jump:
0:00 - Turning slop into gold
2:03 - The trust pyramid and codebase quality
5:00 - Dune and enforceable architecture
6:56 - Replacing comments with lint rules
9:40 - Runtime verification and feature maps
10:50 - Formal verification and UI testing
11:51 - Spending tokens on verification
13:32 - 11,000 PRs and a 0.5% revert rate
15:11 - Foundations before software factories
Watch the full thing.
Then look at your own agent workflow and ask where the real constraints are missing.
Follow @codewithimanshu for more high-signal content that actually moves your career forward.
11,000 pull requests in one month.
Revert rate around 0.5%.
Most AI coding is pure slop.
Lauren Tan’s team at SpaceXSI just showed how they turned it into something that actually ships clean.
Free talk. From the people running Grok Bot agents at scale.
Save it. Watch tonight.
Changes how you think about agents writing code.
You’ve been letting agents dump code and hoping it holds up. This is the opposite approach.
Follow @codewithimanshu for more high-signal content that actually moves your skills forward.
↓
Lauren Tan (poteto).
She talks about the trust pyramid and codebase quality first.
Not “just prompt better.”
They built the foundations so agents can move fast without everything falling apart.
Architecture that enforces itself. Lint rules instead of comments. Tests that actually catch stuff.
Follow @codewithimanshu , I break down talks like this.
↓
Turning slop into gold.
The whole point of the talk.
Most people scale agents and quality dies. More PRs, more bugs, more reverts, more pain.
They did 11k PRs and barely anything got reverted.
That’s not luck. That’s systems.
Runtime verification. Feature maps. Formal verification with TLA+ and Lean for the hard parts.
They spend tokens on checking the work, not just generating more of it.
↓
Cursor agents + pstack.
She shows how they turn the boring engineering foundations into actual software factories.
Agents don’t just write code. They work inside rules that make bad code hard to land.
Dune for enforceable architecture. Replacing “please don’t do this” comments with lint rules that block it.
This is the part most teams skip. They jump straight to agents and wonder why the codebase turns into mess.
More breakdowns of this kind of stuff on @codewithimanshu.
↓
Foundations before software factories.
She keeps coming back to this.
You can’t just throw agents at a messy codebase and expect magic.
Build the constraints first. Make the architecture enforce itself. Verify at runtime and formally where it matters.
Then the agents can go fast and the revert rate stays near zero.
11,000 PRs. 0.5% reverts. That’s the result of doing the boring work first.
↓
Timestamps if you want to jump:
0:00 - Turning slop into gold
2:03 - The trust pyramid and codebase quality
5:00 - Dune and enforceable architecture
6:56 - Replacing comments with lint rules
9:40 - Runtime verification and feature maps
10:50 - Formal verification and UI testing
11:51 - Spending tokens on verification
13:32 - 11,000 PRs and a 0.5% revert rate
15:11 - Foundations before software factories
Watch the full thing.
Then look at your own agent workflow and ask where the real constraints are missing.
Follow @codewithimanshu for more high-signal content that actually moves your career forward.
Thariq vibe coded a GTA 6 level fighting game then sat there unsure how his own characters even worked.
At CODE Summit 2026 he breaks down exactly why that happens to almost everyone.
And how to stop it.
Free talk from the guy who's actually thinking hard about this stuff.
Save it. Watch tonight. Changes how you use coding agents.
You've been letting AI write code without keeping the creative control.
This fixes that.
Follow @codewithimanshu for more high-signal content that actually moves your skills forward.
↓
Thariq Shihipar.
He talks about the Bitter Lesson, Bash, React, MCP.
And the difference between AI's jagged frontier and the smooth interior.
Most people only see the jagged part and get stuck.
He shows the other side.
Follow @codewithimanshu , I break down talks like this.
↓
When vibe coding loses the developer.
This is the part that hit hard.
He built the fighting game with AI.
Then realized he didn't understand how his own characters worked anymore.
That's the trap.
You move fast. You ship. But you lose the mental model of what you built.
Most people never notice until something breaks and they can't fix it without the AI.
Follow @codewithimanshu for more on keeping control while using agents.
↓
Readable specifications + custom scripting languages.
This is his actual solution.
Don't let the agent touch the core engine code.
Separate it.
Write readable specs the agent can work from.
Build a simple scripting language for the game parts.
Then the agent can help without owning the whole thing.
You stay in control. You still understand it.
This is what he demos live.
↓
Live demo: building a scripting language for a game.
He shows the IDE workflow.
How to keep the specs human readable.
How the agent works inside that structure instead of free-for-all generating everything.
It's practical. Not theory.
You can actually copy parts of this approach today.
More breakdowns of talks like this on @codewithimanshu.
↓
Software renaissance idea.
He ties it back to the industrial revolution comparison.
AI can speed things up massively.
But only if we design the workflows so humans still own the understanding and creative decisions.
Otherwise we just get faster at building stuff we don't fully get.
That's the real point of the talk.
Follow @codewithimanshu for more high-signal AI engineering content.
↓
Timestamps if you want to jump around:
0:00 - The sweet side of the Bitter Lesson
1:27 - Rethinking Bash, React and AI art
4:04 - Keeping AI adoption human centered
4:40 - Industrial revolution and software renaissance
5:53 - The jagged frontier and smooth interior
9:19 - Visualizing model knowledge
10:38 - When vibe coding loses the developer
12:54 - Separating engine code from readable specs
13:47 - Live demo: a scripting language for a game
16:40 - Designing a software renaissance
Speaker: Thariq
Watch the full thing. Then try the spec + scripting approach on something small this week.
Follow @codewithimanshu for more high-signal content that actually moves your career forward.
The ideas in my notes are getting made.
Dreamina’s REAL Unlimited top models offer is live:
30 days Seedance 2.5
60 days GPT Image 2.5
Credits no longer set the limit
#DreaminaPartner
Saying goodbye to the Los Santos trio at sunset before hopping in the pickup truck with the new Vice City duo actually hurts 😭🌅
Btw, CapCut is doing 90% off your first month for Black Friday right now if you wanna make your own GTA 6 edits 👇
https://t.co/ESjVv24XWc
#GTA6 #CapCutDidThat
Thariq vibe coded a GTA 6 level fighting game then sat there unsure how his own characters even worked.
At CODE Summit 2026 he breaks down exactly why that happens to almost everyone.
And how to stop it.
Free talk from the guy who's actually thinking hard about this stuff.
Save it. Watch tonight. Changes how you use coding agents.
You've been letting AI write code without keeping the creative control.
This fixes that.
Follow @codewithimanshu for more high-signal content that actually moves your skills forward.
↓
Thariq Shihipar.
He talks about the Bitter Lesson, Bash, React, MCP.
And the difference between AI's jagged frontier and the smooth interior.
Most people only see the jagged part and get stuck.
He shows the other side.
Follow @codewithimanshu , I break down talks like this.
↓
When vibe coding loses the developer.
This is the part that hit hard.
He built the fighting game with AI.
Then realized he didn't understand how his own characters worked anymore.
That's the trap.
You move fast. You ship. But you lose the mental model of what you built.
Most people never notice until something breaks and they can't fix it without the AI.
Follow @codewithimanshu for more on keeping control while using agents.
↓
Readable specifications + custom scripting languages.
This is his actual solution.
Don't let the agent touch the core engine code.
Separate it.
Write readable specs the agent can work from.
Build a simple scripting language for the game parts.
Then the agent can help without owning the whole thing.
You stay in control. You still understand it.
This is what he demos live.
↓
Live demo: building a scripting language for a game.
He shows the IDE workflow.
How to keep the specs human readable.
How the agent works inside that structure instead of free-for-all generating everything.
It's practical. Not theory.
You can actually copy parts of this approach today.
More breakdowns of talks like this on @codewithimanshu.
↓
Software renaissance idea.
He ties it back to the industrial revolution comparison.
AI can speed things up massively.
But only if we design the workflows so humans still own the understanding and creative decisions.
Otherwise we just get faster at building stuff we don't fully get.
That's the real point of the talk.
Follow @codewithimanshu for more high-signal AI engineering content.
↓
Timestamps if you want to jump around:
0:00 - The sweet side of the Bitter Lesson
1:27 - Rethinking Bash, React and AI art
4:04 - Keeping AI adoption human centered
4:40 - Industrial revolution and software renaissance
5:53 - The jagged frontier and smooth interior
9:19 - Visualizing model knowledge
10:38 - When vibe coding loses the developer
12:54 - Separating engine code from readable specs
13:47 - Live demo: a scripting language for a game
16:40 - Designing a software renaissance
Speaker: Thariq
Watch the full thing. Then try the spec + scripting approach on something small this week.
Follow @codewithimanshu for more high-signal content that actually moves your career forward.
@AravSrinivas yes, i feel it does works better with pplx computer.. you should make a comparison graph to show how it's bringing the best results when being used in computer..
Auto-apply to AI-matched real-time jobs with a tailored resume in 2 clicks.
→ https://t.co/vDQEY9M5nv
→ Shows jobs from across the internet in 1 place
.
Join My Community→ https://t.co/s8NPsoA69F
The ideas in my notes are getting made.
Dreamina’s REAL Unlimited top models offer is live:
30 days Seedance 2.5
60 days GPT Image 2.5
Credits no longer set the limit
#DreaminaPartner
Don’t stop at the first concept that comes to mind. Follow the next idea, try a new direction, and explore what’s possible with Dreamina.
#DreaminaCanvas
The recording lets you see the generation process in context. Follow the steps on screen, then imagine how you might adapt the idea with a different prompt or creative direction.
Thirteen years in Los Santos wrapped up in 30 seconds—now it is time for Vice City.
Jump into GPT-6.2 Astra on the CapCut plugin today 🚀
#CapCut62Astra
After the rough cut loaded up, I just fine-tuned the pacing between the final hug on the beach and the two discs sitting side by side in the truck bed.