Gergely Orosz
@gergely.pragmaticengineer.com
Writing The Pragmatic Engineer (@pragmaticengineer.com), the #1 technology newsletter on Substack. Author of The Software Engineer's Guidebook (engguidebook.com). Formerly at Uber, Skype, Skyscanner. More at pragmaticengineer.com
Update: ChatGPT, Claude and Perplexity were all wrong in their interpretation
There is plenty of value to go to a legal professional - heck, to work with a professional in any area that is high stakes! AI sounds smart but it can (+ does) get lots wrong while sounding confident
There is plenty of value to go to a legal professional - heck, to work with a professional in any area that is high stakes! AI sounds smart but it can (+ does) get lots wrong while sounding confident
I’m asking for legal advice for a complicated situation.
Feels like I’m waiting weeks (while a partner reviews the case) and paying a large sum to confirm what Claude and ChatGPT analysed + on the situation and articles is correct, and get it in writing + stamped
Feels like I’m waiting weeks (while a partner reviews the case) and paying a large sum to confirm what Claude and ChatGPT analysed + on the situation and articles is correct, and get it in writing + stamped
January 8, 2026 at 8:39 AM
Update: ChatGPT, Claude and Perplexity were all wrong in their interpretation
There is plenty of value to go to a legal professional - heck, to work with a professional in any area that is high stakes! AI sounds smart but it can (+ does) get lots wrong while sounding confident
There is plenty of value to go to a legal professional - heck, to work with a professional in any area that is high stakes! AI sounds smart but it can (+ does) get lots wrong while sounding confident
Models suddenly getting good enough to write most of my code - which I am now prompting - creates complicated feelings.
It took a long time to get good at coding. And it's not easy. Plus, there was something special about being in "the zone."
Full: blog.pragmaticengineer.com/the-grief-wh...
It took a long time to get good at coding. And it's not easy. Plus, there was something special about being in "the zone."
Full: blog.pragmaticengineer.com/the-grief-wh...
January 7, 2026 at 6:57 PM
Models suddenly getting good enough to write most of my code - which I am now prompting - creates complicated feelings.
It took a long time to get good at coding. And it's not easy. Plus, there was something special about being in "the zone."
Full: blog.pragmaticengineer.com/the-grief-wh...
It took a long time to get good at coding. And it's not easy. Plus, there was something special about being in "the zone."
Full: blog.pragmaticengineer.com/the-grief-wh...
"As part of these updates, {product you pay for} may use customer content and service outputs for internal quality improvement and internal model training purposes.
You may opt out at any time by..."
Don't want to opt out. I want to choose whether to opt in when paying!
You may opt out at any time by..."
Don't want to opt out. I want to choose whether to opt in when paying!
January 7, 2026 at 5:09 PM
"As part of these updates, {product you pay for} may use customer content and service outputs for internal quality improvement and internal model training purposes.
You may opt out at any time by..."
Don't want to opt out. I want to choose whether to opt in when paying!
You may opt out at any time by..."
Don't want to opt out. I want to choose whether to opt in when paying!
On 1 Dec 2020, AWS S3 announced that all writes to S3 are now strongly consistent (not eventually consistent, like before) *at no price change or latency changes* to any customers.
Pulling this off was probably one of the biggest invisible engineering achievements of the decade
Pulling this off was probably one of the biggest invisible engineering achievements of the decade
January 6, 2026 at 9:04 PM
On 1 Dec 2020, AWS S3 announced that all writes to S3 are now strongly consistent (not eventually consistent, like before) *at no price change or latency changes* to any customers.
Pulling this off was probably one of the biggest invisible engineering achievements of the decade
Pulling this off was probably one of the biggest invisible engineering achievements of the decade
We’re entering the age of AI slop that people believe en masse.
This post is 100% fake and probably AI generated. All made up. Yet massive number of upvotes, views and shares.
Journalist @caseynewton.bsky.social got in touch with the “whistleblower.” The guy faked all “evidence” with AI…
This post is 100% fake and probably AI generated. All made up. Yet massive number of upvotes, views and shares.
Journalist @caseynewton.bsky.social got in touch with the “whistleblower.” The guy faked all “evidence” with AI…
January 6, 2026 at 8:03 AM
We’re entering the age of AI slop that people believe en masse.
This post is 100% fake and probably AI generated. All made up. Yet massive number of upvotes, views and shares.
Journalist @caseynewton.bsky.social got in touch with the “whistleblower.” The guy faked all “evidence” with AI…
This post is 100% fake and probably AI generated. All made up. Yet massive number of upvotes, views and shares.
Journalist @caseynewton.bsky.social got in touch with the “whistleblower.” The guy faked all “evidence” with AI…
Reposted by Gergely Orosz
How tech companies measure the impact of AI on software development
It's a hard thing to measure, fwiw. The best summary I know of comes from @lauratacho.com in this deepdive of how 15+ tech companies including Google, GitHub and others do it:
newsletter.pragmaticengineer.com/p/how-tech-c...
newsletter.pragmaticengineer.com/p/how-tech-c...
How tech companies measure the impact of AI on software development
How do GitHub, Google, Dropbox, Monzo, Atlassian, and 13 other companies know how well AI tools work for devs? A deepdive sharing exclusive details, with CTO Laura Tacho
newsletter.pragmaticengineer.com
January 5, 2026 at 10:18 AM
How tech companies measure the impact of AI on software development
From a startup CTO: "Over the break, I realized how good these models have gotten. This year we'll be using them ~10x more vs last year.
So my first order of business: I want to find a solution to measure our *current* dev productivity so I have a baseline to compare to, later."
So my first order of business: I want to find a solution to measure our *current* dev productivity so I have a baseline to compare to, later."
January 5, 2026 at 10:01 AM
From a startup CTO: "Over the break, I realized how good these models have gotten. This year we'll be using them ~10x more vs last year.
So my first order of business: I want to find a solution to measure our *current* dev productivity so I have a baseline to compare to, later."
So my first order of business: I want to find a solution to measure our *current* dev productivity so I have a baseline to compare to, later."
One thing every tech company is doing but few share with anyone:
evaluating AI tools for devs - for coding, for infra (eg gateway), for code review etc. Tons of vendors, unclear which one to buy.
Share what you / your team found, and I'll share what others are seeing and what they measure (cont'd)
evaluating AI tools for devs - for coding, for infra (eg gateway), for code review etc. Tons of vendors, unclear which one to buy.
Share what you / your team found, and I'll share what others are seeing and what they measure (cont'd)
January 4, 2026 at 6:51 PM
One thing every tech company is doing but few share with anyone:
evaluating AI tools for devs - for coding, for infra (eg gateway), for code review etc. Tons of vendors, unclear which one to buy.
Share what you / your team found, and I'll share what others are seeing and what they measure (cont'd)
evaluating AI tools for devs - for coding, for infra (eg gateway), for code review etc. Tons of vendors, unclear which one to buy.
Share what you / your team found, and I'll share what others are seeing and what they measure (cont'd)
I’m asking for legal advice for a complicated situation.
Feels like I’m waiting weeks (while a partner reviews the case) and paying a large sum to confirm what Claude and ChatGPT analysed + on the situation and articles is correct, and get it in writing + stamped
Feels like I’m waiting weeks (while a partner reviews the case) and paying a large sum to confirm what Claude and ChatGPT analysed + on the situation and articles is correct, and get it in writing + stamped
January 4, 2026 at 10:02 AM
I’m asking for legal advice for a complicated situation.
Feels like I’m waiting weeks (while a partner reviews the case) and paying a large sum to confirm what Claude and ChatGPT analysed + on the situation and articles is correct, and get it in writing + stamped
Feels like I’m waiting weeks (while a partner reviews the case) and paying a large sum to confirm what Claude and ChatGPT analysed + on the situation and articles is correct, and get it in writing + stamped
For the last ~20 years, I did most of my coding inside an IDE - the last ~15 with increasingly good autocomplete.
Which is why it’s so weird that I barely opened an IDE the last two weeks, even as I pushed lots of code. I use the CLI, the web and my phone (!!) to prompt code
Which is why it’s so weird that I barely opened an IDE the last two weeks, even as I pushed lots of code. I use the CLI, the web and my phone (!!) to prompt code
January 3, 2026 at 1:59 PM
For the last ~20 years, I did most of my coding inside an IDE - the last ~15 with increasingly good autocomplete.
Which is why it’s so weird that I barely opened an IDE the last two weeks, even as I pushed lots of code. I use the CLI, the web and my phone (!!) to prompt code
Which is why it’s so weird that I barely opened an IDE the last two weeks, even as I pushed lots of code. I use the CLI, the web and my phone (!!) to prompt code
Amazing: The Software Engineer’s Guidebook now has 6 translations, and 8 different paperback versions:
Top (left to right): Japanese, Simplified Chinese (China). Mongolian, Korean
Bottom: English (India edition), English hardcover, German, Traditional Chinese (Taiwan)
Top (left to right): Japanese, Simplified Chinese (China). Mongolian, Korean
Bottom: English (India edition), English hardcover, German, Traditional Chinese (Taiwan)
December 30, 2025 at 11:05 AM
Amazing: The Software Engineer’s Guidebook now has 6 translations, and 8 different paperback versions:
Top (left to right): Japanese, Simplified Chinese (China). Mongolian, Korean
Bottom: English (India edition), English hardcover, German, Traditional Chinese (Taiwan)
Top (left to right): Japanese, Simplified Chinese (China). Mongolian, Korean
Bottom: English (India edition), English hardcover, German, Traditional Chinese (Taiwan)
Two legends confirmed for The Pragmatic Summit: Martin Fowler (@martinfowler.com) and Kent Beck (@kentbeck.com)
We'll talk about what past booms+busts taught them, and their take on AI+software engineering: what they see working, and what not so much
11 Feb, SF: www.pragmaticsummit.com
We'll talk about what past booms+busts taught them, and their take on AI+software engineering: what they see working, and what not so much
11 Feb, SF: www.pragmaticsummit.com
December 29, 2025 at 8:23 PM
Two legends confirmed for The Pragmatic Summit: Martin Fowler (@martinfowler.com) and Kent Beck (@kentbeck.com)
We'll talk about what past booms+busts taught them, and their take on AI+software engineering: what they see working, and what not so much
11 Feb, SF: www.pragmaticsummit.com
We'll talk about what past booms+busts taught them, and their take on AI+software engineering: what they see working, and what not so much
11 Feb, SF: www.pragmaticsummit.com
There's a new trend on LinkedIn where people with zero original thoughts have AI generate what they assume are next-level posts.
Except the AI hallucinates.
The whole thing is nonsense.
Everyone rolls eyes -except the person posting it, thinking "wow, AI makes this easy!"
Except the AI hallucinates.
The whole thing is nonsense.
Everyone rolls eyes -except the person posting it, thinking "wow, AI makes this easy!"
December 23, 2025 at 8:10 PM
There's a new trend on LinkedIn where people with zero original thoughts have AI generate what they assume are next-level posts.
Except the AI hallucinates.
The whole thing is nonsense.
Everyone rolls eyes -except the person posting it, thinking "wow, AI makes this easy!"
Except the AI hallucinates.
The whole thing is nonsense.
Everyone rolls eyes -except the person posting it, thinking "wow, AI makes this easy!"
Whenever I see a prediction on how AI will result in fewer hours worked, I think of AI startups:
They have no budget limit on how much devs can spend on AI (so they spend a TON)
And yet, their devs tend to work MORE than anywhere else though... to outcompete other AI startups!!
They have no budget limit on how much devs can spend on AI (so they spend a TON)
And yet, their devs tend to work MORE than anywhere else though... to outcompete other AI startups!!
December 23, 2025 at 8:28 AM
Whenever I see a prediction on how AI will result in fewer hours worked, I think of AI startups:
They have no budget limit on how much devs can spend on AI (so they spend a TON)
And yet, their devs tend to work MORE than anywhere else though... to outcompete other AI startups!!
They have no budget limit on how much devs can spend on AI (so they spend a TON)
And yet, their devs tend to work MORE than anywhere else though... to outcompete other AI startups!!
Why Revolut is winning and traditional banks left behind:
Last time my “traditional” credit card # got stolen: I noticed it, spent ~15 minutes on the phone to report it; got a new card 1 week later
With Revolut: THEY noticed it, 1 tap to cancel; 1 minute for a new card! Amazing
Last time my “traditional” credit card # got stolen: I noticed it, spent ~15 minutes on the phone to report it; got a new card 1 week later
With Revolut: THEY noticed it, 1 tap to cancel; 1 minute for a new card! Amazing
December 22, 2025 at 10:28 AM
Why Revolut is winning and traditional banks left behind:
Last time my “traditional” credit card # got stolen: I noticed it, spent ~15 minutes on the phone to report it; got a new card 1 week later
With Revolut: THEY noticed it, 1 tap to cancel; 1 minute for a new card! Amazing
Last time my “traditional” credit card # got stolen: I noticed it, spent ~15 minutes on the phone to report it; got a new card 1 week later
With Revolut: THEY noticed it, 1 tap to cancel; 1 minute for a new card! Amazing
My most awkward job offer: when Skyscanner offered a contract, I had to request a weird amendment to it. Independent of my hiring (done by a different division), Skyscanner acquired my brother's startup the same month...
Screenshot from the podcast: newsletter.pragmaticengineer.com/p/design-fir...
Screenshot from the podcast: newsletter.pragmaticengineer.com/p/design-fir...
December 21, 2025 at 8:43 PM
My most awkward job offer: when Skyscanner offered a contract, I had to request a weird amendment to it. Independent of my hiring (done by a different division), Skyscanner acquired my brother's startup the same month...
Screenshot from the podcast: newsletter.pragmaticengineer.com/p/design-fir...
Screenshot from the podcast: newsletter.pragmaticengineer.com/p/design-fir...
What was the Dotcom Boom like? Bryan Cantrill (@bcantrill.bsky.social - cofounder and CTO at Oxide):
"One of my early life lessons: that boom will go on longer than you think possible, and when it switches, it will collapse faster than you can fathom."
"One of my early life lessons: that boom will go on longer than you think possible, and when it switches, it will collapse faster than you can fathom."
December 21, 2025 at 7:56 PM
What was the Dotcom Boom like? Bryan Cantrill (@bcantrill.bsky.social - cofounder and CTO at Oxide):
"One of my early life lessons: that boom will go on longer than you think possible, and when it switches, it will collapse faster than you can fathom."
"One of my early life lessons: that boom will go on longer than you think possible, and when it switches, it will collapse faster than you can fathom."
Probably the most entertaining -and unlikely!- story in tech this year:
A software engineer tricking more than a dozen Y Combinator startups: acing the interview, then delivering little to no work, collecting paychecks, and swindling the next company newsletter.pragmaticengineer.com/p/the-10x-ov...
A software engineer tricking more than a dozen Y Combinator startups: acing the interview, then delivering little to no work, collecting paychecks, and swindling the next company newsletter.pragmaticengineer.com/p/the-10x-ov...
December 20, 2025 at 5:34 PM
Probably the most entertaining -and unlikely!- story in tech this year:
A software engineer tricking more than a dozen Y Combinator startups: acing the interview, then delivering little to no work, collecting paychecks, and swindling the next company newsletter.pragmaticengineer.com/p/the-10x-ov...
A software engineer tricking more than a dozen Y Combinator startups: acing the interview, then delivering little to no work, collecting paychecks, and swindling the next company newsletter.pragmaticengineer.com/p/the-10x-ov...
One really, really, really good use case for AI and coding:
Writing unit integration tests.
I always hated doing these, because most of the effort was about the setup (remembering how to use the test framework, how to create a fake, a mock, the syntax)
Love handing it off!!
Writing unit integration tests.
I always hated doing these, because most of the effort was about the setup (remembering how to use the test framework, how to create a fake, a mock, the syntax)
Love handing it off!!
December 20, 2025 at 3:19 PM
One really, really, really good use case for AI and coding:
Writing unit integration tests.
I always hated doing these, because most of the effort was about the setup (remembering how to use the test framework, how to create a fake, a mock, the syntax)
Love handing it off!!
Writing unit integration tests.
I always hated doing these, because most of the effort was about the setup (remembering how to use the test framework, how to create a fake, a mock, the syntax)
Love handing it off!!
It's interesting to see different experiments on what to do with so much more AI code generated - with much less effort?
Share the prompt itself to code reviewers?
Meta is experimenting with this - good on them! (we need to experiment more IMO)
Share the prompt itself to code reviewers?
Meta is experimenting with this - good on them! (we need to experiment more IMO)
Interesting AI coding feature rolled out at Meta called "trajectories." On diffs, devs can see the prompts used to generate it (if it was AI-generated.) Rolled out to everyone.
Given more code is generated prompting: interesting experiment! Full: newsletter.pragmaticengineer.com/p/the-pulse-...
Given more code is generated prompting: interesting experiment! Full: newsletter.pragmaticengineer.com/p/the-pulse-...
December 19, 2025 at 8:53 AM
It's interesting to see different experiments on what to do with so much more AI code generated - with much less effort?
Share the prompt itself to code reviewers?
Meta is experimenting with this - good on them! (we need to experiment more IMO)
Share the prompt itself to code reviewers?
Meta is experimenting with this - good on them! (we need to experiment more IMO)
Confirmed: Laura Tacho (@lauratacho.com ) will deliver the keynote for The Pragmatic Summit on how AI+coding agents are changing engineering productivity. 11 Feb 2026, SF.
Laura is one the top experts on AI + developer productivity: no fluff, just things that work, and things that don't
(cont'd)
Laura is one the top experts on AI + developer productivity: no fluff, just things that work, and things that don't
(cont'd)
December 18, 2025 at 7:13 PM
Confirmed: Laura Tacho (@lauratacho.com ) will deliver the keynote for The Pragmatic Summit on how AI+coding agents are changing engineering productivity. 11 Feb 2026, SF.
Laura is one the top experts on AI + developer productivity: no fluff, just things that work, and things that don't
(cont'd)
Laura is one the top experts on AI + developer productivity: no fluff, just things that work, and things that don't
(cont'd)
Interesting: I am hearing both inside Meta and inside Google AI coding models simply don't work nearly as well as they do outside:
Models work great on greenfield projects and when using standard tooling.
Both Meta and Google have monoliths and non-standard tooling!
Models work great on greenfield projects and when using standard tooling.
Both Meta and Google have monoliths and non-standard tooling!
December 18, 2025 at 12:19 PM
Interesting: I am hearing both inside Meta and inside Google AI coding models simply don't work nearly as well as they do outside:
Models work great on greenfield projects and when using standard tooling.
Both Meta and Google have monoliths and non-standard tooling!
Models work great on greenfield projects and when using standard tooling.
Both Meta and Google have monoliths and non-standard tooling!
Devs who write about their work/company; talk at conferences - lots of companies in the past used to not care much about this and not value it the least.
Feels like the tide is changing: people who do this but are not valued get hunted down by companies who understand this value
Feels like the tide is changing: people who do this but are not valued get hunted down by companies who understand this value
December 18, 2025 at 8:31 AM
Devs who write about their work/company; talk at conferences - lots of companies in the past used to not care much about this and not value it the least.
Feels like the tide is changing: people who do this but are not valued get hunted down by companies who understand this value
Feels like the tide is changing: people who do this but are not valued get hunted down by companies who understand this value
How have servers and the cloud evolved in the last 30 years, and what might be next? @bcantrill.bsky.social has been at the thick of the industry since the Dotcom Boom, and shares fascinating stories.
Bryan is one of my all-time favorite people to talk with - don't miss this one.
(cont'd)
Bryan is one of my all-time favorite people to talk with - don't miss this one.
(cont'd)
December 17, 2025 at 9:02 PM
How have servers and the cloud evolved in the last 30 years, and what might be next? @bcantrill.bsky.social has been at the thick of the industry since the Dotcom Boom, and shares fascinating stories.
Bryan is one of my all-time favorite people to talk with - don't miss this one.
(cont'd)
Bryan is one of my all-time favorite people to talk with - don't miss this one.
(cont'd)
More details than shared before about the us-east-1 outage in October.
No, it was not caused by “brain drain” (an assumption that those building the services left) - creators of the service causing the issue (DNS Enactor) were on the outage call!
Distributed systems are hard
No, it was not caused by “brain drain” (an assumption that those building the services left) - creators of the service causing the issue (DNS Enactor) were on the outage call!
Distributed systems are hard
December 16, 2025 at 6:54 PM
More details than shared before about the us-east-1 outage in October.
No, it was not caused by “brain drain” (an assumption that those building the services left) - creators of the service causing the issue (DNS Enactor) were on the outage call!
Distributed systems are hard
No, it was not caused by “brain drain” (an assumption that those building the services left) - creators of the service causing the issue (DNS Enactor) were on the outage call!
Distributed systems are hard