Twitxy Saved

Matt Stockton

Verified account This account is verified. Learn more

@mstockton

Mostly tweeting LLM stuff. Working w/ companies to help them do AI and ML @ pragmanexus.com.
Writing thoughts here: mattstockton.com/writing.html
Milwaukee, WI mattstockton.com/writing.html Joined January 2009
2.8K Following 12K Followers 3.5K Posts

Latest posts 5

Pinned post

Matt Stockton Verified account @mstockton
By far my biggest advice to anyone trying to adopt AI properly:

1. Pay a little bit of money to Anthropic
2. Download Claude Code
3. Open Claude Code
4. Press 'Shift-Tab' until it says 'plan mode on'
5. Open Voice Memo on your iPhone. Just talk about all the things you want to accomplish. When you think you are done, just keep talking. Make sure it is at least 10 minutes, hopefully longer
6. Send this Voice Memo to your computer
7. Download MacWhisper and use it to transcribe this voice memo. Trust me, you will want MacWhisper and will use it later a lot
8. Type into Claude Code: "I have never used you before but I talked about some things. I will paste those things in below. Please read the things and ask me any questions you need to in order to help me figure out how to use you to be awesome. Ask me lots of questions until I tell you I am done"
9. Then paste in the transcript
10. Then press enter

Then just let Claude take the wheel, and them please send me a DM if this works.

Also, if this just sounds crazy, just literally take this entire message and paste it into whatever AI you are using and say 'some weird person told me to paste this into you, I want to use it, but I don't know how. What should I do?'

I am just trying to help you get started. Curiosity and persistence are the most important things.
360 replies 1.2K reposts 17K likes 2.4M views View on X
Matt Stockton Verified account @mstockton
Good read here. Many Eng teams are trying to figure out patterns for pragmatic AI code review and there’s no easy answer. You need to be intentional w/ how you architect it. This idea finds a nice balance where you try to let things flow through ‘automatically’ thus increasing your velocity, without losing confidence in your code quality, or degrading to sloppy ai reviews because your team is exhausted (eg agents fighting agents)

Basically, a set of classification rules for a change set, and if the change matches those rules, it can either be approved or ignored (no comments allowed) - plus a post processing step where, if a bad change was let through, the team needs to adapt and modify the classification rules.

Lots of way to do these things, but you need to be intentional about it.
Will Larson Verified account @Lethain Oct 17
Wrote up how we're trying to reduce PR review friction with concept of "small-change" as PRs/eng/week has increased 10x on some teams. (Aside: incidents down 42% HoH, critical incidents down 78%! Velocity vs quality never been a falser dichotomy.)

lethain.com/small-changes/
View quoted post on X
9 replies 2 reposts 75 likes 14K views View on X
Matt Stockton Verified account @mstockton
US Senator on intelligence committee with some thoughts on root cause for outages. Basically - companies patching things because they have access to better models and have been told what’s coming. Kind of have to do it in a rush. This is honestly a best case root cause
Mark Warner Verified account @MarkWarner 22h
I understand why Americans are concerned about the widespread outages we’ve seen in recent days, affecting everything from banks and major technology companies to critical government systems. At this point, there is no indication that these incidents are the result of a cyberattack, foreign interference, or some coordinated effort.

What appears to be happening instead is a consequence of how quickly artificial intelligence is changing the cybersecurity landscape.

AI is making it possible to identify vulnerabilities across enormous numbers of systems at unprecedented speed, creating enormous pressure on companies and government agencies to patch those vulnerabilities immediately, often all at once, putting tremendous strain on systems that weren’t built to handle that pace of change.

While we still need to get to the bottom of these incidents, the evidence so far points toward legitimate efforts to secure our systems, not malicious activity.

Still, these disruptions should serve as a wake-up call. It feels like every day brings another reminder that AI is moving faster than our ability to manage its consequences.

This is exactly why I’ve been pushing for commonsense guardrails on AI and stronger cybersecurity protections.

Unfortunately, President Trump has shown little interest in confronting these challenges, his administration has repeatedly undermined its own credibility, and its cuts to CISA have weakened the very expertise we need to navigate moments like this.

Americans deserve clear, reliable information and a government that is prepared to respond. I’m going to keep digging into these incidents, pressing for answers, and making sure we’re doing everything possible to protect our critical systems. But based on what we know today, there is good reason to believe these disruptions are the unintended consequences of efforts to make our systems safer, not evidence of a broader attack.
View quoted post on X
2 replies 0 reposts 9 likes 2.1K views View on X
Matt Stockton Verified account @mstockton
This is good. Although fully down the rabbit-hole, I've still held the core belief that these tools are *accessible to all* - but are exponential multipliers for people who 'know what good looks like' - often, this takes the form of constraints you prescribe to the agent around architecture, patterns, etc. - I think this does still matter a ton. Opus 5.5 / Fable 5.2, etc. are amazing, but steering still helps a lot.

I do think the attached argument is directionally correct though, in that eventually these 'steering constraints' become sub-optimal for the models, and you'd get a better result by staying quiet. Bitter lesson for sure.

Key quote from the linked article:

"Just as in ML, the reason the lesson is bitter is that knowledge being displaced is the expertise we’ve spent our careers acquiring and the expertise we pride ourselves makes us “senior” and “great engineers”. We’re comfortable automating pure implementation, but considerably less comfortable with the thought that our architectural judgment would be not a force multiplier, but a constraint to better outcomes. Like the linguists of the early LLM era, we may find our own intuition to be inferior to machine intelligence."
Jacob Lauritzen Verified account @jacsebl 1d
Engineers naturally want to encode their own clever ideas and domain insights into systems design. That's provides personal satisfaction and better outcomes today, but will soon be inferior to letting agents discover the right abstractions themselves x.com/i/article/2122…
View quoted post on X
2 replies 2 reposts 26 likes 3.7K views View on X