iiinova — where your digital self keeps evolving

Joe Fabisevich @mergesort.me · 06/10/2026

More proof that you should never publicly make a prediction.

The
dows 95
mIs
1995 Blog
The Year the Fu
Prediction of the year, 1995: Internet
'will soon go spectacularly supernova'
The Internet's emergence into mainstream consciousness in 1995
was met no small amount of skepticism and eye-rolling. It was all
overhyped, critics said. A distraction.
Notable among the skeptics was astrophysicist Clifford Stoll
whose 1995 book, Silicon Snake Oil, offered such predictions
about the digital world as:
"I don't believe that phone books, newspapers, magazines, or
corner video stores will disappear as computer networks spread.
Nor do I think that my telephone will merge with my computer,
to become some sort of information appliance."
And this howler: "Video-on-demand, that killer application of
communications, will remain a dream."
But the most spectacular, off-the-mark
prediction of the year 1995 had to be that of Bob
Metcalfe, the multimillionaire inventor of
Ethernet technology and founder of 3Com
Corporation. He also was publisher of
InfoWorld, for which he wrote a weekly column
Metcalfe (UT-Austin)
called "From the Ether."
In the issue of Infoworld dated December 4, 1995 - 20 years
ago tomorrow - Metcalfe wrote that the Internet "will soon go
spectacularly supernova and in 1996 will collapse." He identified

Joe Fabisevich @mergesort.me · 06/10/2026

I'm sorry but this is called reading a book. Or a blog post, the internet is full of amazing content. Or go to an intellectually stimulating event in your area. It's not that hard, you don't have to invent a thinking gym.

Andrew Yeung w
@andruyeung
10/4/26, 10:53 PM
We created gyms because modern day
work no longer required physical activity.
Without exercise, our muscles atrophy.
I predict we'll need gyms for our brains
too, once Al starts doing our knowledge
work.
• 807
9392
[7 1025

7562

Joe Fabisevich @mergesort.me · 05/10/2026

That's some real good obfuscation there Zoom.

1240

Joe Fabisevich @mergesort.me · 05/10/2026

Not only did it build a two apps to repro both issues, record a video, and write a radar draft, but the app itself provides a reviewer impossible to mess up repro steps.

010

Joe Fabisevich @mergesort.me · 05/10/2026

Going forward this is the only way I will file annoying to reproduce radars.

| want you to resume our work where we left off. For context please
read our AGENTS.md, then temp/app-intents/README.md and its
linked investigation documents. Supporting evidence is now under
temp/app-intents/evidence/ .
Our plan was to continue working where we left off, but instead | want
you to review the work we've been doing to try and fix our broken App
Intents. We have a lot of debug code that we've written that we
should not commit to production, a few staged changes, and a lot of
unstaged changes.
Our new goal is to file a radar for the code that isn't working, with the
most isolated and reproducible code that we can provide. To do this
we want to keep the unstaged changes that are not relevant to
Applintents untouched. We should keep the staged code because it
contains Appintents correctness fixes. But in our unstaged changes,
we have a lot of Shortcuts code that we've been working on that still
does not work, and rather than continuing to investigate we ought to
be pulled out into this radar.
We need to create a new project that demonstrates our bug. We need
to be able to reproduce it in this new project, and only once we can
reproduce it 10x in a row (using @ Device Hub) should we consider it
reproducible. That is when we can consider removing it from Plinky,
but | do not want you to remove any code or make any commits on
your own.
| want you to create a full reproduction of this in a small project I've
created called Shortcuts Boolean Bug, in ~/Developer. | want you to
build the minimal scaffolding needed and run through the complete
workflow, attempting to reproduce this bug. Don’t give up if your first
attempt fails, instead | want you to continue trying until you succeed
at reproducing the bug with as little code as we can — providing a
good bug report.
Then | want you to capture a screen recording of the simulator so we
can submit a video showing the issue, and write up a first draft of a
radar that will show the issue to an Apple engineer.

140

Joe Fabisevich @mergesort.me · 04/10/2026

Looking at New York from Vermont across Lake Champlain and it couldn’t be any more beautiful.

0130

Joe Fabisevich @mergesort.me · 01/10/2026

Happy first day of spooky season 2026 to all!

A black cat with bright eyes lies on a grey blanket with white shapes, yawning widely, revealing bright teeth. The scene is softly lit, with a window showing a dark sky in the background.

171

Joe Fabisevich @mergesort.me · 28/09/2026

I've never put in less effort to deliver high-quality working products, while at the same time having more unfinished side projects than ever. This post does a great job explaining why that schism can occur, since the side projects rarely reflect *my* care or values. seangoedecke.com/human-ai-par...

That doesn't mean I can leave the Al
alone. Purely vibe-coding at work
produces awful outputs. But they're not
awful because they're bad code, they're
awful because they're in bad taste: code
that is not maintainable, that trades off
important requirements in order to satisfy
made-up ones, that contradicts the long-
term strategy for a feature or service, and
so on.
In other words, my primary value is not
that I help the Al write better code, it's
that I align the Al with the values of my
organization. Human-Al partnerships
are for alignment, not capability.

Frontier models are misaligned to the
working programmer. They are obsessed
with a set of behaviors that presumably
satisty their RL grader: writing enormous
block comments above functions,
producing hundreds of useless unit tests,
adding little bits of text all over websites
they design, and so on. Working with
agents is about noticing and wrestling
with those behaviors. That's why my
prompting advice is to explicitly talk
about your high-level values: it's an
attempt to head off obvious
misalignment.

Alignment is also more context-
dependent than capability. Working code
is working code, no matter what (which is
partially why it's comparatively easy to
train for). But aligning to a company's
technical values is different from
company to company, as any software
engineer who's switched companies
knows. It can almost feel like relearning
the job. So training an aligned coding
model doesn't just require hitting the
exact right set of values, it requires
creating a model that can adapt on the
fly to a wide range of possible values.

240

Joe Fabisevich @mergesort.me · 26/09/2026

TFW you’re a little demon, but you’re also a pretty girl. #caturday

A black cat sitting in front of some roses

1213

Joe Fabisevich @mergesort.me · 25/09/2026

Those two photos were from the same spot, facing different directions. And as I walk down the street it seems like everyone in NYC is out right now staring at the sky. Really big fan of this weird spooky energy.

A cityscape at dusk, with a deep purple and orange sky filled with wispy clouds. In the foreground, train tracks stretch across the scene, and buildings are silhouetted against the glowing horizon. A sign reads, “NEW YORK CITY.”

090

Joe Fabisevich @mergesort.me · 25/09/2026

I have no idea what's going on with the sky right now, but it's a real vibe up there.

A low-angle view of tall buildings lining a sidewalk at dusk. The sky is pink and purple, and parked cars are visible beside the path. A sign reads, "BAY VIEW."

A tilted street scene at sunset, with a bright orange and yellow sky and a tall pole. Cars are parked along the curb, and buildings line the road. A traffic light shows "YELLOW".

370

Joe Fabisevich @mergesort.me · 25/09/2026

So can I, big whoop Claude.

020

Joe Fabisevich @mergesort.me · 23/09/2026

Of course, IKEA was always such a natural fit for bench marking.

Epoch Al
@EpochAIResearch
9/23/26, 1:21 PM Typefully
Can Al tell if you've built your IKEA
furniture wrong? Our new benchmark, the
Furniture Assembly Benchmark (FAB),
gives models the manual and a photo of a
half-completed piece of furniture and asks
them to spot the mistake.
The top score has gone from 28% to 80%
in just 10 months.
Al has improved at spotting errors in furniture
assembly
Open Al
• Anthropic
• Moonshot
Google
Accuracy
100%
Alibaba
80%
GPT-6 Astra •
Fable 5.10
60%
8~ GPT-5.6 Sol
GPT-5.5
40%
GPT-5.2
Opus 4.5
20%
0%

1457

Joe Fabisevich @mergesort.me · 23/09/2026

Wow, posted this dumb joke before I even saw the Claude video. Source code here: github.com/JohnHeibel/PDoomVideo

000

Joe Fabisevich @mergesort.me · 23/09/2026

Where's Waldo, Maiori edition. She's in there somewhere.

A black cat hidden in plain sight against a black TV

141

Joe Fabisevich @mergesort.me · 22/09/2026

Never has a website been more targeted at me.

A blue-gray banner with white text reads, "Like Joseph – First Rejected, Then Revealed." The text is set against a dark blue background. Below, a gray browser window displays the URL "jews.com"

020

Joe Fabisevich @mergesort.me · 18/09/2026

Hooray, Anthropic finally gave up!

Thariq C
@trq212
9/18/26, 2:04 PM
We're adding support for AGENTS.md to
Claude Code.
Starting today in version 2.1.277, if there is
no CLAUDE.md in a folder, Claude will
check for and use AGENTS.md.
You can toggle this behavior in /config.
368
3562
7 610 •

832521

Joe Fabisevich @mergesort.me · 18/09/2026

I made this prediction before AI 2027, before Anthropic released Claude Code, and before we had anything resembling agents. All I needed was to try o1 to know the world was going to change, fast, and that it was going to be messy.

Joe Fabisevich
mergesort
Terrifying to consider that in all
likelihood we will reach AGI during
Donald Trump's presidency.
Mar 4, 2025 at 9:22 AM

241

Joe Fabisevich @mergesort.me · 18/09/2026

I am a very serious director who writes very serious docs to help improve the state of our organization's security.

1Password also stores Passkeys. Nobody knows what
Passkeys are, but every website wants you to login with one.
(Ok, ok, | know what they are. Imagine a website asks you to
Sign In With Google, but you don’t need Google — instead
you sign in with your Password Manager)

170

Joe Fabisevich @mergesort.me · 17/09/2026

💯💯💯. Wrote this in my going away email at Twitter, but we had an opportunity to become the central identity layer of the internet and chose not to. Now I'm rooting for ATProto to do this but better, since it can be built from first principles. fabisevi.ch/2022/04/01/goodby...

000

Joe Fabisevich @mergesort.me · 15/09/2026

Me: Whatever you do, don't do a Skynet. Them: AI-first war baby, awoo! (Wolf howl 🐺)

A black-and-white image with bold lettering and Donald Trump raising an arm in a salute. The background is dark with faded stars and stripes. The words “AI-FIRST WAR DEPARTMENT” are written in bold, jagged letters.

020

Joe Fabisevich @mergesort.me · 14/09/2026

It's my favorite nerd day of the year, new iOS day! I'm a bit bummed to not ship day one updates for Plinky this year, but I had a hard time getting them in after starting my new job. I've got a bunch in the works though, so for now I'll have to enjoy all the great updates from my favorite indies.

5:39 Ø
•ll 5GW
89
<
App Updates
Things 3
Today
Open
Things is ready for iOS 27 and watchOS 27,
with great new system integrations:
• Full integration with the new Siri (please
check if it's available on your device).
• Full integration with Spotlight, your
device's built-in search.
• More snoozing options for notifications,
with more suitable durations.
• A new extra-tall widget to see longer lists.
• Various other compatibility improvements
for the new OS.
See our blog for more information:
www.things.app
Version 3.24 • 64.7 MB
Flighty - Live
Flight Tracker
Today
Open
A refined Dynamic Island, flight answers
from Siri, and two new looks for your Home
Screen.
# DYNAMIC ISLAND REFINEMENTS
See how long until departure with a
countdown that now appears even in the

0110

Joe Fabisevich @mergesort.me · 12/09/2026

Come to Copenhagen for the New Orleans jazz parades. 🎶

031

Joe Fabisevich @mergesort.me · 12/09/2026

Now this is how you immortalize your pet.

1160

Joe Fabisevich @mergesort.me · 10/09/2026

The first thing I've noticed about Copenhagen is how well designed everything is. Even something as simple as a food label that bolds allergens solves a real problem that people have in a very elegant manner.

Balls of chocolate in a container, with an ingredient list

150

Joe Fabisevich @mergesort.me · 09/09/2026

Never knew Picasso was inspired by macOS.

A platter that looks like the Finder icon

040

Joe Fabisevich @mergesort.me · 08/09/2026

OpenAI's agents opening up HuggingFace.

Elizabeth Holmes saying:
WHY WOULD I DECEIVE YOU?

070

Joe Fabisevich @mergesort.me · 07/09/2026

Happy Labor Day from this postcard view.

040

Joe Fabisevich @mergesort.me · 07/09/2026

I won't be around for my league's fantasy draft, so I had ChatGPT aggregate a bunch of expert's draft boards to put together an auto-draft for my team. I feel a little weird offloading this task, but the alternative is having Yahoo's shitty engine do it for me, so I'm curious to see how this goes.

Work
Worked for 16m 36s ›
Done-the 10-team auto-draft setup is
saved in Yahoo.
• 190 players ranked in the exact
planned order
• QB/TE timing, sleepers, and
demotions applied
• Kickers and defenses reserved for
late rounds
• Tyreek Hill, Ricky Pearsall, and Jayden
Higgins marked Do Not Draft
• Reloaded Yahoo and confirmed all
settings persisted with zero ranking
mismatches
We can revise it once the final team count
is known.

130

Joe Fabisevich @mergesort.me · 05/09/2026

Decided to spend Labor Day weekend in a country that really understands how to do labor. 🇫🇷

A tropical garden scene in the French Riviera

1100

Joe Fabisevich @mergesort.me · 04/09/2026

OpenAI out here just inventing new tiers of thinking.

2100

Joe Fabisevich @mergesort.me · 04/09/2026

When Claude designs your brand you get a company called Fish Audio with a mammal for the logo.

If this biliboard
could talk it'd
sound exactly
like you.
fish.audio

Voice Al. Minus the Al voice.
Fish Audio

4241

Joe Fabisevich @mergesort.me · 02/09/2026

Emo cat staring at her pile of toys.

141

Joe Fabisevich @mergesort.me · 31/08/2026

Use Meta's new Muse Code to build your own vibecoded ass website.

160

Joe Fabisevich @mergesort.me · 31/08/2026

Waking up several times hurt your total? Yeah, well, maybe you can deal with the screaming cat next time.

OK Sleep Score

You scored a 72 last night. Waking up several times hurt your total.

160

Joe Fabisevich @mergesort.me · 29/08/2026

In their defense they weren't trying to pull one over on anyone, they posted the same exact thing in their reply at the same time they posted the first tweet.

0100

Joe Fabisevich @mergesort.me · 29/08/2026

The highest form of compliment you can give me.

A scene from Pretty Little Liars that says: Your software
slays by the way.

190

Joe Fabisevich @mergesort.me · 29/08/2026

If there's one company that's actually not all ears…

Warby Parker
11:30 AM >
We're all ears
Tell us about your experience with us.

051

Joe Fabisevich @mergesort.me · 28/08/2026

So close to hitting publish on this essay… must resist the urge to write a giant thread instead. 😵‍💫

A Slight Tangent On Compounding
Errors
A compounding error occurs when the output
of one mistaken step becomes an input or
assumption for subsequent steps. People
view systemic errors as independent, but
more often they're correlated across similar
environments.
When you see a coworker give their two
weeks notice your initial assumption is that
the person left on their own accord. When
you see three people quit in a month, it's less
likely to be an outlier situation and more likely
that it's a trend stemming from the new new
manager who's giving off bad vibes.
In navigation if a ship were to chart a course
that's one degree off, the error may seem
negligible but will be only grow over time.
Every mile traveled increases the error, and by
the time you notice there's you have a gaping
chasm between where you are and where you
set out to be.

The smallest of mistakes can lead to large
consequences, and the earlier in the process
this happens the worse the outcomes will be.
You may be making the best decision every
step of the way, but if your initial premise is
off then you shouldn't expect good results.
Every time your agent makes a mistake it's
not just one mistake, but the foundation of
future mistakes that leave you with more
problems than you started with.

110

Joe Fabisevich @mergesort.me · 27/08/2026

I bet on offering high quality support because even when you don't make a sale, you can get a person to change their mind. ☺️

120

Joe Fabisevich @mergesort.me · 25/08/2026

They all laugh at you until @plinky.app is the #1 out of 1 link-saving apps advertised on streaming TV.

020

Joe Fabisevich @mergesort.me · 25/08/2026

I know this is wrong because I'm on the N right now, but it's scary that this is a totally plausible wait time for the N train.

040

Joe Fabisevich @mergesort.me · 24/08/2026

Followed me into the bathroom demanding to be pet. Wouldn't take no for an answer so she settled into the sink until I gave her five straight minutes of pets and purring.

1131