Husband to wonderfull wife, JW, software engineering manager @GEICO, ex-Microsoft, like helping people. Posts and thoughts are my own.
Los Angeles, CA
Joined September 2016
- Tweets1.9K
- Following264
- Followers127
- Likes5.8K
I’m speaking at AGNTCon + MCPCon North America 2026!
Shipping Agent Skills Safely: CI/CD, Evals, and Guardrails
📅 Oct 23 • 2:20–2:45 PM PT 📍 Room 210B, Concourse Level
1
2
45
Sessions:
events.linuxfoundation.org/a…
Livestream:
cvent.com/api/email/dispatch…
26
"Welcome back, Mr. and Mrs. Smith! My agent told me it's your wedding anniversary... Wait, didn't you celebrate your anniversary here last week???"
a legit problem in tech is how many of its leaders are trying to reinvent systems without bothering to understand how or why the current system works in the first place
79
This is making a lot of assumptions. Our systems grow in complexity everyday. One of the reasons why bugs happen is because we couldn't foresee all the ways the code would be used. Even smarter models can fix that.
41
Honest question, why release a demo of the product where I, the human, can barely understand the audio, but the captions on X got most of it?
Here's a demo of a conversation between two people at Deveillance @be_inaudible
Watch the live transcription fall apart.
44
Because the overwhelming cost of software isn't in creating it; it's in maintenance.
Unpopular opinion:
Software should get cheaper.
If AI lets us build in 2 days what used to take 2 months, why are we still pricing it like nothing changed?
We can’t have 10x productivity without prices eventually coming down.
37
Sure we can generate a bunch of things with AI but when everyone can do it, they what makes this special? In addition looking at this it's clearly fake and breaking laws of physics.
Nike spends $2M on a stadium shoot for a 10-second hero shot.
40-person crew. 3 days of permits. 2 weeks in post. Stunt doubles. Insurance paperwork thicker than the script.
This creator generated the entire sequence on a laptop for under $80.
A guy sprints across a stadium roof at night, leaps off the edge, free-falls into a sold-out arena, and lands on a giant football in the center of the pitch.
One continuous shot. No crane. No drone operator. No safety harness review board.
The toolchain behind this clip:
> Kimi K3 wrote the full shot sequence: rooftop sprint, jump arc, mid-air hang, landing impact, crowd reaction timing
> Kling 3.0 generated the fluid body physics and night lighting on the roof
> Seedance 2.5 rendered the free-fall, wind on the shirt, and the ball deformation on impact
> ElevenLabs produced the ambient stadium atmosphere and crowd noise
> CapCut handled speed ramps, color grade, and final export
A traditional production house would need a stadium rental at $150K, a stunt coordinator, a medical team on standby, and 6 months of liability clearance.
This entire sequence was prompted, rendered, and exported between dinner and sleep.
AI didn't replace the stunt double.
It replaced the entire stadium booking.
1
40
I always allow duplication to happen then extract what really is the pattern. DRY needs to be balanced with other concerns. Should be treated as a principle not law.
I think “never repeat yourself” has damaged more codebases than duplication ever did.
Developers see the same five lines twice and immediately invent a generic abstraction with 14 parameters.
Sometimes repeating yourself is cheaper than being clever.
3
41
LinkedIn Open to Work helped me get a job after I lost my job at Microsoft. Its an easy way to let people in your network know your looking. I was also hiring manager and never looked down on anyone with Open to Work or employment gaps on their resume.
Tip: never use the green Open to Work banner on LinkedIn
Negative signal for hiring managers and recruiters
2
3
131
Michael Larson retweeted
NEW: malware developers added nuclear & biological weapons text to to their spyware.
Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner.
Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky.
When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit.
We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted.
In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation.
H/T to colleagues that shared this with me socket.dev/blog/mini-shai-hu…
Michael Larson retweeted
my instagram (@ korn) was stolen overnight via the Meta AI exploit and was subsequently disabled.
it was Meta Verified, facial scan verified, and had 0 TOS violations.
the account is the sole source of my income.
i spent 6 hours trying to get human support and meta's support AI gave me 4 broken links in a row. we're at the point where one AI stole it and another can't fix it, zero humans in the loop anywhere.
does anyone know how to get in contact with meta support about this?
it is absolutely ridiculous that a trillion dollar company allows an exploit like this and doesn't even address it.
@instagram @AIatMeta @DarkWebInformer
Exactly @ThePrimeagen, if it's just a matter of scaling agents, we should all start up our own swarm of millions of agents so we can all be trillionaires.
Honestly why stop at 100x engineer?
Just use more agents, you literally could be 1000x, 10000x, 100000x just by scaling
You could what you use to in an entire year in one second
92
We have two cats. I like to think we are one happy family. My wife shattered my illusion today by telling me "Dogs have owners, cats have staff." 😅
44
Putting in you Claude.md file "DO NOT DELETE THE PRODUCTION DATABASE!!!!" Doesn't work. What does? The safeguards that have prevented production issues before AI was a thing. I explore that in my latest Substack post.
open.substack.com/pub/mlarso…
90
Michael Larson retweeted
Warning to open source maintainers: the Axios supply chain attack started with some
very sophisticated social engineering targeted at one of their developers simonwillison.net/2026/Apr/3…