Allen Holub
@allenholub.bsky.social
Author, international speaker, consultant, software architect, kitchen-sink wrangler.
A user story is not a code word for specification or a requirement or any description of work. It is not a ticket. It is not a to-do item. It is not something you can build.
1/5
1/5
February 9, 2026 at 9:04 PM
A user story is not a code word for specification or a requirement or any description of work. It is not a ticket. It is not a to-do item. It is not something you can build.
1/5
1/5
I've been using the term "AI-assisted software engineering" as a counter to "vibe coding," which I see as something else entirely. The term itself is a mouthful, though. Can anybody recommend something more pithy?
February 9, 2026 at 7:11 PM
I've been using the term "AI-assisted software engineering" as a counter to "vibe coding," which I see as something else entirely. The term itself is a mouthful, though. Can anybody recommend something more pithy?
No metric improves product decisions. You do that by talking to your customers.
What does improve product decisions is empathy, feedback, and understanding the market.
What does improve product decisions is empathy, feedback, and understanding the market.
February 9, 2026 at 6:39 PM
No metric improves product decisions. You do that by talking to your customers.
What does improve product decisions is empathy, feedback, and understanding the market.
What does improve product decisions is empathy, feedback, and understanding the market.
I keep reading that quality needs to take a back seat to profit, but I'm beginning to think that the people who say that don't define "quality" the same way I do.
Let's take "well architected" as an example.
1/9
Let's take "well architected" as an example.
1/9
February 9, 2026 at 6:03 PM
I keep reading that quality needs to take a back seat to profit, but I'm beginning to think that the people who say that don't define "quality" the same way I do.
Let's take "well architected" as an example.
1/9
Let's take "well architected" as an example.
1/9
Demanding an estimate guarantees the estimate will not be useful to the business and will always drive costs up.
Software engineers all know that estimates are useless.
1/6
Software engineers all know that estimates are useless.
1/6
February 6, 2026 at 7:15 PM
Demanding an estimate guarantees the estimate will not be useful to the business and will always drive costs up.
Software engineers all know that estimates are useless.
1/6
Software engineers all know that estimates are useless.
1/6
I'm not a fan of story points; I don't use or recommend them.
That said, they can have some value inside the team when used as originally intended. The original XP team invented story points to obfuscate time estimates from management.
1/5
That said, they can have some value inside the team when used as originally intended. The original XP team invented story points to obfuscate time estimates from management.
1/5
February 5, 2026 at 7:59 PM
I'm not a fan of story points; I don't use or recommend them.
That said, they can have some value inside the team when used as originally intended. The original XP team invented story points to obfuscate time estimates from management.
1/5
That said, they can have some value inside the team when used as originally intended. The original XP team invented story points to obfuscate time estimates from management.
1/5
“Too many GPUs make you lazy.” One way to address AI energy usage is to improve (and specialize) the models. Mistral is a good example of that: t.ly/6dTxd
February 5, 2026 at 6:01 PM
“Too many GPUs make you lazy.” One way to address AI energy usage is to improve (and specialize) the models. Mistral is a good example of that: t.ly/6dTxd
The AI/LLM haters are legion, but they have a very real concern about the environmental cost of the tech. To say that all LLMs burn down the planet is simplistic, however. We have to evaluate the vendors and models and choose based on their commitment to clean energy.
1/11
1/11
February 4, 2026 at 6:56 PM
The AI/LLM haters are legion, but they have a very real concern about the environmental cost of the tech. To say that all LLMs burn down the planet is simplistic, however. We have to evaluate the vendors and models and choose based on their commitment to clean energy.
1/11
1/11
Maybe the third time's a charm 😄. Yesterday, I posted that we could eliminate things like country-name dropdowns and replace them with free text, perhaps using an LLM to correct errors.
1/8
1/8
February 2, 2026 at 11:05 PM
Maybe the third time's a charm 😄. Yesterday, I posted that we could eliminate things like country-name dropdowns and replace them with free text, perhaps using an LLM to correct errors.
1/8
1/8
So, a few weeks ago, I posted about those ridiculous country-name drop-downs with 100 or more lines and similar 50-line ones for US state codes. (Don't get me started on year-picker dropdowns with 100+ sequential numbers in them 🙄.
1/6
1/6
February 1, 2026 at 5:40 PM
So, a few weeks ago, I posted about those ridiculous country-name drop-downs with 100 or more lines and similar 50-line ones for US state codes. (Don't get me started on year-picker dropdowns with 100+ sequential numbers in them 🙄.
1/6
1/6
I keep reading posts about people who wrote applications entirely with "zero coding." They say they don't read the code and instead refine the application using increasingly precise prompts. I have some questions:
* Is the application nontrivial?
1/7
* Is the application nontrivial?
1/7
January 31, 2026 at 6:32 PM
I keep reading posts about people who wrote applications entirely with "zero coding." They say they don't read the code and instead refine the application using increasingly precise prompts. I have some questions:
* Is the application nontrivial?
1/7
* Is the application nontrivial?
1/7
If we need to invest $206,756.00 in 40 Mac minis (fully loaded), something's seriously wrong with the way we do our work.
January 26, 2026 at 9:23 PM
If we need to invest $206,756.00 in 40 Mac minis (fully loaded), something's seriously wrong with the way we do our work.
I've noticed that the new way trolls discount ideas that make them uncomfortable is by accusing me of being an LLM. Since these comments almost always refer to me as "that guy" or "buddy," I can't help but wonder if they are LLM-generated 😄.
January 26, 2026 at 8:14 PM
I've noticed that the new way trolls discount ideas that make them uncomfortable is by accusing me of being an LLM. Since these comments almost always refer to me as "that guy" or "buddy," I can't help but wonder if they are LLM-generated 😄.
Reposted by Allen Holub
High school students using AI to write college application essays that are then graded by AI from admissions departments is a great example of how AI usage devalues systems while people can accurately point to “productivity gains” from using AI.
January 26, 2026 at 2:04 AM
High school students using AI to write college application essays that are then graded by AI from admissions departments is a great example of how AI usage devalues systems while people can accurately point to “productivity gains” from using AI.
A few days back, I wrote about how AI can be useful, and that we ignore or reject it at our peril. Paradoxically, as several people pointed out, some not very politely, longer-term productivity in many (maybe most) shops that use AI code generation has gone down. That's absolutely true.
1/8
1/8
January 26, 2026 at 7:59 PM
A few days back, I wrote about how AI can be useful, and that we ignore or reject it at our peril. Paradoxically, as several people pointed out, some not very politely, longer-term productivity in many (maybe most) shops that use AI code generation has gone down. That's absolutely true.
1/8
1/8
How do you measure productivity? Let’s start by talking about the engineering obsession with quantitative metrics. You can easily measure productivity on an assembly line. The faster a car comes off the line at a given quality level, the more productive you are.
1/10
1/10
January 25, 2026 at 7:25 PM
How do you measure productivity? Let’s start by talking about the engineering obsession with quantitative metrics. You can easily measure productivity on an assembly line. The faster a car comes off the line at a given quality level, the more productive you are.
1/10
1/10
This image is a perfect example of everything that is wrong with the usual code-review process. People debate the grammar, even though there's no universe in which the prepositional phrase "of cell phones and earbuds" can be the subject of a sentence.
1/9
1/9
January 23, 2026 at 5:53 PM
This image is a perfect example of everything that is wrong with the usual code-review process. People debate the grammar, even though there's no universe in which the prepositional phrase "of cell phones and earbuds" can be the subject of a sentence.
1/9
1/9
Any software manufacturer that forces its customers into email-based 2FA is losing its customers. Get with the program.
January 22, 2026 at 7:16 PM
Any software manufacturer that forces its customers into email-based 2FA is losing its customers. Get with the program.
<rant> I'm getting tired of AI being the boogie man. The problem is not AI. The problem is the people who use AI to do reprehensible things.
AI is not taking away anybody's job. The people who run the companies that use AI are taking away people's jobs.
1/10
AI is not taking away anybody's job. The people who run the companies that use AI are taking away people's jobs.
1/10
January 21, 2026 at 10:22 PM
<rant> I'm getting tired of AI being the boogie man. The problem is not AI. The problem is the people who use AI to do reprehensible things.
AI is not taking away anybody's job. The people who run the companies that use AI are taking away people's jobs.
1/10
AI is not taking away anybody's job. The people who run the companies that use AI are taking away people's jobs.
1/10
Reposted by Allen Holub
Judging by the comments on my most recent AI-related post, the terms "AI" and "The boogie man" have become synonyms. What happened to nuance?
January 21, 2026 at 8:33 PM
Judging by the comments on my most recent AI-related post, the terms "AI" and "The boogie man" have become synonyms. What happened to nuance?
Discounting a technology or process because the evidence is "anecdotal" is a straw man. There are no studies for most practices & tools, and when there are, the numbers are flawed. For example, you cannot measure productivity using PRs or KLOC. Generating lots of garbage quickly is not productive.
January 21, 2026 at 6:54 PM
Discounting a technology or process because the evidence is "anecdotal" is a straw man. There are no studies for most practices & tools, and when there are, the numbers are flawed. For example, you cannot measure productivity using PRs or KLOC. Generating lots of garbage quickly is not productive.
This is a common argument against AI. All the AI does is create code, which is the easiest part of software engineering. Humans are the architects. Humans assure the quality. It's like building a house with unskilled labor that requires supervision. No thinking is outsourced. LLMs can't think.
AI assisted software engineering is a horrible idea.
It's like building a house by asking a fortune telling automaton.
you outsource critical parts of your thinking and analysis, give up crucial abilities, and end up with houses that cave in on people.
Eventually.
It's like building a house by asking a fortune telling automaton.
you outsource critical parts of your thinking and analysis, give up crucial abilities, and end up with houses that cave in on people.
Eventually.
January 21, 2026 at 4:55 PM
This is a common argument against AI. All the AI does is create code, which is the easiest part of software engineering. Humans are the architects. Humans assure the quality. It's like building a house with unskilled labor that requires supervision. No thinking is outsourced. LLMs can't think.
Discounting a technology or process because the evidence is "anecdotal" (i.e., people's actual experience) is a straw man. There are no quant studies, & when there are, the numbers are flawed. E.g., you can't measure productivity with PRs or KLOC. Creating lots of garbage quickly is not productive.
January 21, 2026 at 4:47 PM
Discounting a technology or process because the evidence is "anecdotal" (i.e., people's actual experience) is a straw man. There are no quant studies, & when there are, the numbers are flawed. E.g., you can't measure productivity with PRs or KLOC. Creating lots of garbage quickly is not productive.
Of the critical responses to this post that weren't just invective and name-calling (I've blocked those particular people), one important observation was that people complained about having to deal with AI-generated garbage that other people checked into the repo. 1/
AI-assisted software engineering is like the flush toilet 😄
I just read an article about "dumb phones." There is a class of people who have decided that smartphones are evil, so they've returned to phone technology circa 1990. No apps. No big screen or QWERTY keyboard. Just a phone (and txt).
1/10
I just read an article about "dumb phones." There is a class of people who have decided that smartphones are evil, so they've returned to phone technology circa 1990. No apps. No big screen or QWERTY keyboard. Just a phone (and txt).
1/10
January 21, 2026 at 2:38 AM
Of the critical responses to this post that weren't just invective and name-calling (I've blocked those particular people), one important observation was that people complained about having to deal with AI-generated garbage that other people checked into the repo. 1/