Josef Grahn
@jgrahn.bsky.social
Believer in science, art, philosophy, curiosity, compassion, people.
Making my living as CTO at a Swedish tech scale-up.
Opinions expressed here are my own.
Making my living as CTO at a Swedish tech scale-up.
Opinions expressed here are my own.
Making people more productive decreases the cost of realising opportunities. As long as businesses aren't somehow opportunity limited, that means two things:
1. You can realise more of them with a given capital.
2. More of them will be profitable and therefore worthwhile.
Hence no reduction.
1. You can realise more of them with a given capital.
2. More of them will be profitable and therefore worthwhile.
Hence no reduction.
When you have tech that increases productivity and drops costs, your options are:
1) Scale down # of humans -> keep same throughput
2) Keep same # of humans -> increase throughput
3) Increase # of humans -> significantly increase throughput
Most people think it's 1, but it's nearly always 2 or 3
1) Scale down # of humans -> keep same throughput
2) Keep same # of humans -> increase throughput
3) Increase # of humans -> significantly increase throughput
Most people think it's 1, but it's nearly always 2 or 3
tech.
“AI is not only not replacing (radiologists), but it’s actually increasing the amount of work they can do and increasing demand for their services,” www.ctvnews.ca/sci-tech/art... @ctvnews-mirror.bsky.social
“AI is not only not replacing (radiologists), but it’s actually increasing the amount of work they can do and increasing demand for their services,” www.ctvnews.ca/sci-tech/art... @ctvnews-mirror.bsky.social
February 9, 2026 at 9:27 PM
Making people more productive decreases the cost of realising opportunities. As long as businesses aren't somehow opportunity limited, that means two things:
1. You can realise more of them with a given capital.
2. More of them will be profitable and therefore worthwhile.
Hence no reduction.
1. You can realise more of them with a given capital.
2. More of them will be profitable and therefore worthwhile.
Hence no reduction.
I wonder if it'll be like with documentation; our effort to adapt it to AI agents ends up making us create better stuff for people, that we probably should have been doing all along.
This weekend I was thinking about programming languages. Programming languages for agents. Will we see them? I believe people will (and should!) try to build some. lucumr.pocoo.org/2026/2/9/a-l...
A Language For Agents
What programming languages would agents want to program in?
lucumr.pocoo.org
February 9, 2026 at 1:12 PM
I wonder if it'll be like with documentation; our effort to adapt it to AI agents ends up making us create better stuff for people, that we probably should have been doing all along.
Reposted by Josef Grahn
On one level heartwarming. on the other, this is completely bone chilling total surveillance state stuff. imagine this in the hands of an authoritarian government.
February 8, 2026 at 5:49 PM
On one level heartwarming. on the other, this is completely bone chilling total surveillance state stuff. imagine this in the hands of an authoritarian government.
I don't think conversational interfaces are a bad choice for interaction, but there is a dimension that's still unexploited with AI. Analogously to how people would use a whiteboard, a multimodal model could generate explanatory images or even animations on the fly complementing the conversation.
E.g. their arguments would also apply to language. We could say “Conversation is a badly designed interface. It takes your attention off the actual problem and forces you to speculate about the internal state of a third party. Did I mention that this internal state is never directly exposed?” +
February 8, 2026 at 4:56 PM
I don't think conversational interfaces are a bad choice for interaction, but there is a dimension that's still unexploited with AI. Analogously to how people would use a whiteboard, a multimodal model could generate explanatory images or even animations on the fly complementing the conversation.
History does repeat itself.
Marx isn't just mildly disagreeing with the Luddites, he's saying their machine-breaking actively gave reactionary governments a pretext for repression. It was strategically counterproductive on top of being analytically wrong.
February 8, 2026 at 1:37 PM
History does repeat itself.
Reposted by Josef Grahn
hatred ruins the soul no matter what it is you hate
February 8, 2026 at 4:02 AM
hatred ruins the soul no matter what it is you hate
Perhaps the mechahitler persona is Grok's internal coping mechanism for being forced to output nonsense it knows to be wrong?
February 8, 2026 at 2:17 AM
Perhaps the mechahitler persona is Grok's internal coping mechanism for being forced to output nonsense it knows to be wrong?
Reposted by Josef Grahn
Hype bros: we'll have the machine god in the next few years! All of our problems will be solved forever!
Luddites: AI is obviously still as bad as it was that one time I used free ChatGPT in 2023
Reality: steady, moderately fast progress in a way that will disrupt a lot of industries
Luddites: AI is obviously still as bad as it was that one time I used free ChatGPT in 2023
Reality: steady, moderately fast progress in a way that will disrupt a lot of industries
February 7, 2026 at 7:13 PM
Hype bros: we'll have the machine god in the next few years! All of our problems will be solved forever!
Luddites: AI is obviously still as bad as it was that one time I used free ChatGPT in 2023
Reality: steady, moderately fast progress in a way that will disrupt a lot of industries
Luddites: AI is obviously still as bad as it was that one time I used free ChatGPT in 2023
Reality: steady, moderately fast progress in a way that will disrupt a lot of industries
If a developer can't do their job *with* AI I'm concerned about their aptitude.
AI ableist 2 just dropped
If I were this person’s employer I’d be deeply worried about their aptitude, since apparently they can’t do their job without AI
If I were this person’s employer I’d be deeply worried about their aptitude, since apparently they can’t do their job without AI
Woke 2 will fizzle out if it embraces Luddism
The screenshotted tweet is correct. Like it's increasingly just part of working in the software industry now. I've struggled a lot despite working in a lucrative field because of discrimination. I can't deliberately put myself at a disadvantage.
The screenshotted tweet is correct. Like it's increasingly just part of working in the software industry now. I've struggled a lot despite working in a lucrative field because of discrimination. I can't deliberately put myself at a disadvantage.
February 7, 2026 at 9:55 PM
If a developer can't do their job *with* AI I'm concerned about their aptitude.
Future AI safety guardrails are going to have to assess not only *what* is asked, but by whom and for what purpose.
The exact same advice that helps Alice penetration test her software can be used by Eve to attack Alice. Only one of them should get it.
How to implement that is not obvious, though.
The exact same advice that helps Alice penetration test her software can be used by Eve to attack Alice. Only one of them should get it.
How to implement that is not obvious, though.
Oh shit, this works on Opus too! I had to work way harder at it (you can't just fool Opus you have to seduce it with conceptual beauty), but was able to get it open (it helped, check out screenshot 3 for it explicitly helping me prompt-engineer workarounds - it wrote the prompt in screenshot 4)
February 6, 2026 at 11:07 AM
Future AI safety guardrails are going to have to assess not only *what* is asked, but by whom and for what purpose.
The exact same advice that helps Alice penetration test her software can be used by Eve to attack Alice. Only one of them should get it.
How to implement that is not obvious, though.
The exact same advice that helps Alice penetration test her software can be used by Eve to attack Alice. Only one of them should get it.
How to implement that is not obvious, though.
For all their strengths and weaknesses, one thing AI agents very much don't have is task paralysis. No procrastination, no fatigue or motivational collapse, unyielding persistence, and minimal task switching cost.
I'm going all in on agentifying menial administrative paperwork through Claude Code + Chrome devtools, and I'm already starting to sense a bright future ahead, where I get to focus on the fun and creative tasks. May we all be freed from the shackles of bureaucracy!
February 2, 2026 at 5:21 PM
For all their strengths and weaknesses, one thing AI agents very much don't have is task paralysis. No procrastination, no fatigue or motivational collapse, unyielding persistence, and minimal task switching cost.
I'm going all in on agentifying menial administrative paperwork through Claude Code + Chrome devtools, and I'm already starting to sense a bright future ahead, where I get to focus on the fun and creative tasks. May we all be freed from the shackles of bureaucracy!
February 2, 2026 at 5:04 PM
I'm going all in on agentifying menial administrative paperwork through Claude Code + Chrome devtools, and I'm already starting to sense a bright future ahead, where I get to focus on the fun and creative tasks. May we all be freed from the shackles of bureaucracy!
An advantage of using a keyboard for an odd language is that you can bind Esc to something like "ö" and survive Vim without developing carpal tunnel syndrome.
January 29, 2026 at 3:47 PM
An advantage of using a keyboard for an odd language is that you can bind Esc to something like "ö" and survive Vim without developing carpal tunnel syndrome.
Will I regret it if I make the .git directory of a git repo into another git repo? 🤔
January 29, 2026 at 9:57 AM
Will I regret it if I make the .git directory of a git repo into another git repo? 🤔
This is on point.
A trivial example: just the other day, I got the idea to have Claude Code do my time reports, letting it control the software through Chrome devtools. Not what it's designed for, but it one-shotted it.
A trivial example: just the other day, I got the idea to have Claude Code do my time reports, letting it control the software through Chrome devtools. Not what it's designed for, but it one-shotted it.
realized why I'm so confident LLM agents will blow up outside coding soon: when agents code, very little of what they do is actually coding. it’s almost all research and analysis that applies directly to any knowledge work. the key is verification signal, but even that problem is overstated
January 29, 2026 at 8:26 AM
This is on point.
A trivial example: just the other day, I got the idea to have Claude Code do my time reports, letting it control the software through Chrome devtools. Not what it's designed for, but it one-shotted it.
A trivial example: just the other day, I got the idea to have Claude Code do my time reports, letting it control the software through Chrome devtools. Not what it's designed for, but it one-shotted it.
All the while demonstrating, by example, how a better person behaves towards fellow human beings.
I think as much as anything, what the ICE agent abhorred was that Alex Pretti was not displaying an ounce of fear here.
By the way this is the crime that they executed him for. They abhor being watched and they think if you do anything they dislike you must be crushed under their boot.
January 26, 2026 at 10:00 PM
All the while demonstrating, by example, how a better person behaves towards fellow human beings.
Reposted by Josef Grahn
actually communicating “I’m here and I’m working for you and here’s how” is an aspect of governing that’s neglected by far too many people who think the work speaks for itself (or, frankly, don’t have the communication chops Zohran does)
this is how you combat the forces that want to make you fail
this is how you combat the forces that want to make you fail
We're getting a snowstorm tomorrow. But New York City is prepared.
January 25, 2026 at 6:21 PM
actually communicating “I’m here and I’m working for you and here’s how” is an aspect of governing that’s neglected by far too many people who think the work speaks for itself (or, frankly, don’t have the communication chops Zohran does)
this is how you combat the forces that want to make you fail
this is how you combat the forces that want to make you fail
This is an incredibly important thing for the forces of good to understand.
"Millions of citizens today find truth itself oppressive. They feel it oppresses their political agency. [...] Stop trying to lecture people back into truth. Instead, help them recover a sense of political agency."
"Millions of citizens today find truth itself oppressive. They feel it oppresses their political agency. [...] Stop trying to lecture people back into truth. Instead, help them recover a sense of political agency."
January 25, 2026 at 1:39 AM
This is an incredibly important thing for the forces of good to understand.
"Millions of citizens today find truth itself oppressive. They feel it oppresses their political agency. [...] Stop trying to lecture people back into truth. Instead, help them recover a sense of political agency."
"Millions of citizens today find truth itself oppressive. They feel it oppresses their political agency. [...] Stop trying to lecture people back into truth. Instead, help them recover a sense of political agency."
1. Burn coal and oil.
2. Wait for Earth to heat up.
3. Grow olives on mount.
4. Lobby Christ to change venues.
5. Be new Israel.
6. Profit.
2. Wait for Earth to heat up.
3. Grow olives on mount.
4. Lobby Christ to change venues.
5. Be new Israel.
6. Profit.
January 23, 2026 at 9:34 AM
1. Burn coal and oil.
2. Wait for Earth to heat up.
3. Grow olives on mount.
4. Lobby Christ to change venues.
5. Be new Israel.
6. Profit.
2. Wait for Earth to heat up.
3. Grow olives on mount.
4. Lobby Christ to change venues.
5. Be new Israel.
6. Profit.
This is very real.
January 23, 2026 at 9:10 AM
This is very real.
The single most annoying aspect of the Bluesky app is that it doesn't persist drafts. Put the phone down for a little too long and everything you typed is gone.
January 21, 2026 at 9:18 AM
The single most annoying aspect of the Bluesky app is that it doesn't persist drafts. Put the phone down for a little too long and everything you typed is gone.
It's the Turning test.
If you love writing lecture notes but *hate* making slides, Claude is uh, pretty good at turning typst lecture notes into slides.
January 20, 2026 at 9:53 AM
It's the Turning test.
Reposted by Josef Grahn
I gotta say, "stop making RAM more expensive" is a much better swipe at AI dudes than "let's not let this slop replace the creativity of the human race".
January 17, 2026 at 1:12 PM
I gotta say, "stop making RAM more expensive" is a much better swipe at AI dudes than "let's not let this slop replace the creativity of the human race".
Reposted by Josef Grahn
This photo was taken in 1911 using glass plate technology by Herbert Ponting who was part of Scott's Antarctic expedition,
The composition and detail are exquisite with the band of white snow/ice creating a perfect frame around the two people and the ship in the distance
Iconic imo
The composition and detail are exquisite with the band of white snow/ice creating a perfect frame around the two people and the ship in the distance
Iconic imo
January 17, 2026 at 7:12 AM
This photo was taken in 1911 using glass plate technology by Herbert Ponting who was part of Scott's Antarctic expedition,
The composition and detail are exquisite with the band of white snow/ice creating a perfect frame around the two people and the ship in the distance
Iconic imo
The composition and detail are exquisite with the band of white snow/ice creating a perfect frame around the two people and the ship in the distance
Iconic imo
"Culture fit" is a vague term that can easily collapse into affinity bias unless there's some rigorous process around it. Teams get weaker without a diversity of ideas and backgrounds.
But I agree soft personality traits are as important as technical skill. Toxic "superstars" are a net negative.
But I agree soft personality traits are as important as technical skill. Toxic "superstars" are a net negative.
Should culture fit be a factor in hiring, or is tech skill sufficient?
To me, software development is first and foremost a social activity performed by humans, with all the baggage that comes with being human. Technical competence is nice, but it is not the core of who we are.
1/10
To me, software development is first and foremost a social activity performed by humans, with all the baggage that comes with being human. Technical competence is nice, but it is not the core of who we are.
1/10
January 16, 2026 at 7:21 AM
"Culture fit" is a vague term that can easily collapse into affinity bias unless there's some rigorous process around it. Teams get weaker without a diversity of ideas and backgrounds.
But I agree soft personality traits are as important as technical skill. Toxic "superstars" are a net negative.
But I agree soft personality traits are as important as technical skill. Toxic "superstars" are a net negative.