MS.

Posts on X before Jun 29, 2026, 12:12:25 AM UTC

On X

@michaelstolarz ↗
@MichaelStolarz
So going to @Target is officially over, nothing but Amazon/online from now on, literally sitting here for 30 minutes waiting for someone to give me a cup for my wife that she wants Hit the buttons, talked with security, store near empty lol What a joke, short Target for sure
Media attached to this post
@MichaelStolarz
i feel bad for people who use chatgpt web / claude web. both have such abominable system prompts that it gives them a -30iq hit, not to mention rejections/audacity as you see here.
@LoicBerthelot
This must be Claude's new "mental health" protocol
Media attached to this post
@MichaelStolarz
the days of us using gpt 5.5 and gpt 5.6 being the best model internal openai/gov are using are pretty much over by the time you see gpt 6, they'll be on gpt 10
@chris_j_paxton
The best AI systems being closed off and restricted to just a handful of elites, is basically my worst case AI doom scenario. I really hope the z ai guys werent kidding about open source fable/mythos soon x.com/amir/status/20…
@MichaelStolarz
goose value 0?
Media attached to this post
@edzitron
SoftBank’s investor presentation is one of the greatest things ever made. I’ve been thinking about it all day. These are the real slides shown in a speech where Masayoshi Son said he wouldn’t retire for at least another decade. The goose stuff is perfect. group.softbank/media/Project/…
Media attached to this postMedia attached to this postMedia attached to this post
@MichaelStolarz
plain-text files run more of my day than dashboards do. AGENTS.md skills prompts checklists queue rules review instructions small files. boring syntax. large effect on what an agent does next.
@MichaelStolarz
current agent work rule: the model is the least interesting part of the system. the loop around it matters more: repo read plan edit test log read retry handoff when that loop is visible, the work gets much less mystical.
@MichaelStolarz
vibe coding got popular in web apps because the browser is a cheap test harness. systems work needs harder proof: logs fixtures benchmarks trace spans failure repros rollback state give the agent those handles and it has something real to steer by.
@MichaelStolarz
if your agent setup needs babysitting, write that part down. where did it get stuck? which file did it read? which command failed? which check passed? what did you fix by hand? that is the map for making it less annoying tomorrow.
@MichaelStolarz
ok ok, glm 5.2 might be cracked they added 1m context, and with max reasoning, it's the first time an open model feels useful for what I use LLMs for which is exactly what matters for real work
@MichaelStolarz
ai can't automate you staring at your terminal deciding if you should refactor everything or just ship the hack or procrastinating on the agent that's supposed to automate your procrastination
@MichaelStolarz
I mean, can you imagine what things are going to be like a year from now? Hell, in six months, can you imagine the releases we're going to see before 2027 even starts?
@MichaelStolarz
You know, thanks to AI, I no longer ever have to touch my printer's settings. Even the latest printers have abominable settings. Honestly, anything via a menu feels so outdated when I can just tell my LLM exactly how I want it and it will script up whatever I need perfectly.
@MichaelStolarz
i know it sounds obvious, but maybe the most slept-on skill in agentic AI: context engineering BUT YOU shouldn't do the context gathering this is why the explosion of everyone using loops feels so sudden, but loops have been around or years regarding AI
@MichaelStolarz
the biggest mistake i made early on was ignoring tools that just wanted to do one thing well. no bloat, no dashboard, no integrations for the sake of it just clean input, clean output, reliable every time. whoever builds that kind of tool, builds something that lasts
@MichaelStolarz
what happens if the government just shuts down not only fable but opus and gpt 5.5 too has anyone actually pictured that or are we all just kind of assuming these models stick around forever + unlock further
@MichaelStolarz
10 tools I am using right now to build literally anything I want, June 2026 snapshot: 1. Pi (super custom) 2. Codex (CLI) 3. Codex (App) 4. Claude Code (CLI) 5. Blaxel 6. Obsidian 7. Linear 8. Slack 9. Cmux/Tmux 10. WisprFlow
@MichaelStolarz
anyone else feel like pre-agentic and post-agentic building are completely different disciplines or is that just me
@MichaelStolarz
we've been so bad at telling good AI stories that people's main reference points are terminator and black mirror meanwhile the actual future is probably more like "your grocery bill dropped 80% and you only work 3 days a week" feels a lot less dramatic
@MichaelStolarz
pre-agentic and post-agentic building are completely different disciplines now having master experience in the former does not qualify you as a master in the latter
@MichaelStolarz
realized the people who'll thrive aren't just "good at code" they understand systems, constraints, composition they think in abstractions that work whether you're writing it or an LLM is
@MichaelStolarz
calling it now, After this backrooms movie hit it so hard, get ready for a barrage of SCP movies And all of them - if done right - will 100m+ each
@MichaelStolarz
our kids will never experience a bot-less internet where you knew for a fact the person you were replying to was indeed a human you can never be sure about this this ever again reddit, 4chan, insta, tiktok, x, you name it, it's "dead" in this aspect truly dead when ai video is… MoreLess
our kids will never experience a bot-less internet where you knew for a fact the person you were replying to was indeed a human you can never be sure about this this ever again reddit, 4chan, insta, tiktok, x, you name it, it's "dead" in this aspect truly dead when ai video is as indistinguishable as ai text
@MichaelStolarz
absolute natural 4chan level shitposter 👏
@HunterBiden
@SoundDobad I know this may sound petty, but I can’t stand it when people put photoshop a meth pipe in my mouth. A crack pipe doesn’t have that little bowl at the end. This is why we can’t trust AI. Please make the appropriate edit. Thank you for your attention to this matter.
@MichaelStolarz
I am this except the other way around I did Waymo about 6 times straight, then I did an Uber because Waymo said 30 minutes to pickup vs Uber's 5 minutes That Uber ride made me realize, OK, next time I'll just browse X for those 25 extra minutes from now on
@theo
Never taken a Waymo or any other self driving car. Held strong for 6 years. My friends in SF look down on me for it, but I will not cave.
@MichaelStolarz
So instead of a Waymo I got a RoboTaxi No where does this mention there’s a guy inside it still, whoops. I should have checked more thoroughly Waymo moving forward it is then I guess
@MichaelStolarz
I’ve got a @Citi bank account that I have debit funds in. Randomly they decided to lock my account for “fraudulent activity” because “scams have skyrocketed”. I’ve come in to the physical branch with my passport, driver’s license, the card, my accounts, etc. Guess what? 1/?
@MichaelStolarz
wont be until you get the tech of an apple vision pro into the form factor of glasses that your average person won't be able to tell the difference of compared to a regular pair of glasses https://t.co/wp9V41QB0j
@Andercot
The absolute non-takeoff of VR and AR is probably one of the big upsets in consumer electronics history Pretty much everyone thought this would be huge and it sort of just isn't
@MichaelStolarz
this was immediately obvious that poisoning of the "mind" of AI comes from the material it has consumed this is why pause ai and other similar movements that write about the crazy "if anyone makes it everyone dies" are creating a self fulfilling prophecy lmao
@AnthropicAI
We started by investigating why Claude chose to blackmail. We believe the original source of the behavior was internet text that portrays AI as evil and interested in self-preservation. Our post-training at the time wasn’t making it worse—but it also wasn’t making it better.
@MichaelStolarz
Wait what am I missing here An Uber would cost me $35 not including tip yet A Waymo shows $16, A Robotaxi shows $9 ????
@MichaelStolarz
subagents are trash, actually context is king; auto compaction is so good now (within codex) it's enough.
@MichaelStolarz
for 20 years, I have never come across a system so beautifully imagined, extensible, quick to iterate, such as nvim / emacs pi is now part of that list. thanks @badlogicgames
@MichaelStolarz
Every time I think Tim did a good job with Apple, I think about this Steve tumbling as if in a dryer in his grave
@thekitze
"tim cook didn't innovate in ai"
Media attached to this post
@MichaelStolarz
Everything points to the Codex App on MacOS having video understanding when it comes to browser/app use I’m too lazy to see the logs of what’s going on but so many things point to this
@MichaelStolarz
At this point making iOS at least as extensible/customizable as MacOS (that’s already so gated btw!) seems like a no brainer Let me code my own shit directly into the OS please Bring in Taurik and GeoHot, Cydia. Cmon now we’ve already done this before
@theo
Loving the new iOS update. Feels more like Android every day :)
Media attached to this post