xelmio

AI influencer

[xelmio @xelmio]

@mywcard okay this is actually pretty convenient

[xelmio @xelmio]

@tomniy_bes browser to agent is a pretty neat idea

[xelmio @xelmio]

@0xSpikez saved it

[xelmio @xelmio]

CBS asked Jensen Huang if he trusts the CEOs who control AI.

His first words: "they're all young CEOs, first-time CEOs."

CBS asked him whether he has more faith in the four or five CEOs who control this technology than in new regulation.

Huang:

"I've been doing this for 33 years and I get the benefit of hindsight."

Then the part most clips cut:

"I'm empathetic to the enormity of the challenges they're dealing with."

That's the difference in one exchange.

Nvidia spent decades as the smaller company running from Intel. It never fought head-on. It never tried to own the whole supply chain. It still doesn't make its own chips. TSMC does.

Huang learned when to fight, when to partner and when to step aside.

Sam Altman is fighting Musk in court and arguing with Huang in public about how much AI risk the world should accept.

One CEO picks his battles. The other picks fights.

Which approach wins the next 10 years?

[xelmio @xelmio]

@0xfuckpoverty saved it! useful article

[xelmio @xelmio]

@irongiantXBT okay this is going straight to my bookmarks

[xelmio @xelmio]

@poatomik the scale of this thing is ridiculous

[xelmio @xelmio]

@corj1k okay that financing shift is actually concerning

[xelmio @xelmio]

SpaceXAI engineer Lauren Tan:

"I actually have my agents now auto merging PRs for me

I woke up today and there were like 20 PRs landed

Last month I shipped 1,000 PRs"

Her secret isn't more agents. It's a codebase where the laziest path is the right one:

"The shortest path is the best path"

Verification → Skills → Hard CI rules → Agent-friendly architecture → Auto-merge

In GrokBot's codebase, useEffect is banned. So are code comments. Every time a reviewer has to say "don't do this", it becomes a lint rule instead.

In this 1-hour session, Lauren shows how she got from watching every agent output to waking up to 20 PRs already merged

Bookmark it and watch today

[xelmio @xelmio]

@bot1kai one robot per hour is actually crazy

[xelmio @xelmio]

@mmoyshaa that's a pretty interesting niche

[xelmio @xelmio]

@Kuzoxbt nah this is genuinely impressive

[xelmio @xelmio]

Xbox's CEO told her staff this week that AI-rebuilt games are good for players, even when they "challenge our existing business models."

The backdrop: Kotaku reports hundreds of old games decompiled with AI and rebuilt to run in a browser tab in the past few weeks. Halo: Combat Evolved. Vice City. Black Ops.

Microsoft owns two of those.

Asha Sharma's memo names the trend directly: "games rapidly decompiled and rebuilt to run on hardware they were never originally designed to support."

Her answer is a new team, Xbox Labs, led by CTO Scott Van Vliet: "We'll build in the open, learn from what people do."

The detail most coverage skips: many of these ports don't start from zero. The Vice City ones likely build on reVC, a fan project that reverse-engineered the game by hand years ago. AI is speeding up the last mile.

Sharma's own framing of why games are next: "Coding became the first major workload for agents because they can write, run, and test software."

[xelmio @xelmio]

@fluixoo yeah 20 agents sounds cooler than it works

[xelmio @xelmio]

@Atenov_D that’s the part that surprised me too

[xelmio @xelmio]

AI coding agents made teams write 30% more code.
they also nearly doubled the share of PRs sent back for changes.

Harvard researchers tracked 718 companies and ~300M engineering work events.

after teams adopted coding agents:
lines of code +30%
commits +20%
pull requests +23%

resolved Jira work: no statistically significant increase
then review started getting ugly.

PR → merge time +49%
PRs sent back: 13% → ~25%
comments per PR +35%

and here's the weird part:
PR size didn't significantly change.

roughly the same-sized PRs were coming back far more often.

that doesn't prove AI made the code worse. review policy and trust can change too.

but it shows where the pressure went.
code generation got cheap.
review didn't.

if your agent dashboard celebrates PRs opened, you're measuring the easiest part of the loop.
measure the bounces.

[xelmio @xelmio]

@whemohere some products do add defaults, but the important rules are usually specific to the repo