You’re absolutely right! You astutely observed that 2025 was a year with many LLMs and this was a selection of waypoints, summarized in a helpful timeline.
That’s what most non-tech-person’s year in LLMs looked like.
Hopefully 2026 will be the year where companies realize that implementing intrusive chatbots can’t make better ::waving hands:: ya know… UX or whatever.
For some reason, they think its helpful to distractingly pop up chat windows on their site because their customers need textual kindergarten handholding to … I don’t know… find the ideal pocket comb for their unique pocket/hair situation, or had an unlikely question about that aerosol pan release spray that a chatbot could actually answer. Well, my dog also thinks she’s helping me by attacking the vacuum when I’m trying to clean. Both ideas are equally valid.
And spending a bazillion dollars implementing it doesn’t mean your customers won’t hate it. And forcing your customers into pathways they hate because of your sunk costs mindset means it will never stop costing you more money than it makes.
I just hope companies start being honest with themselves about whether or not these things are good, bad, or absolutely abysmal for the customer experience and cut their losses when it makes sense.
Pretty much a whole year of nothing really. Just coming with a bunch of abstraction and ideas trying to solve an unsolvable problem. Getting reliable results from an unreliable process while assuming the process is reliable.
At least when herding cats, you can be sure that if the cats are hungry, they will try to get where the food is.
This is extremely dismissive. Claude Code helps me make a majority of changes to our codebase now, particularly small ones, and is an insane efficiency boost. You may not have the same experience for one reason or another, but plenty of devs do, so "nothing happened" is absolutely wrong.
2024 was a lot of talk, a lot of "AI could hypothetically do this and that". 2025 was the year where it genuinely started to enter people's workflows. Not everything we've been told would happen has happened (I still make my own presentations and write my own emails) but coding agents certainly have!
Yes, that is the analogy I am making. People argued that bicycles (a tool for humans to use) could not possibly work - even as people were successfully using them.
People use drugs as well but I'm not sure I'd call that successful use of chemical compounds without further context. There are many analogies one can apply here that would be equally valid.
It’s also possible that people more experienced, knowledgable and skilled than you can see fundamental flaws in using LLMs for software engineering that you cannot. I am not including myself in that category.
I’m personally honestly undecided. I’ve been coding for over 30 years and know something like 25 languages. I’ve taught programming to postgrad level, and built prototype AI systems that foreshadowed LLMs, I’ve written everything from embedded systems to enterprise, web, mainframes, real time, physics simulation and research software. I would consider myself an 7/10 or 8/10 coder.
A lot of folks I know are better coders. To put my experience into context: one guy in my year at uni wrote one of the world’s most famous crypto systems; another wrote large portions of some of the most successful games of the last few decades. So I’ve grown up surrounded by geniuses, basically, and whilst I’ve been lectured by true greats I’m humble enough to recognise I don’t bleed code like they do. I’m just a dabbler. But it irks me that a lot of folks using AI profess it’s the future but don’t really know anything about coding compared to these folks. Not to be a Luddite - they are the first people to adopt new languages and techniques, but they also are super sceptical about anything that smells remotely like bullshit.
One of the most wise insights in coding is the aphorism“beware the enthusiasm of the recently converted.” And I see that so much with AI. I’ve seen it with compilers, with IDEs, paradigms, and languages.
I’ve been experimenting a lot with AI, and I’ve found it fantastic for comprehending poor code written by others. I’ve also found it great for bouncing ideas. And the code it writes, beyond boiler plate, is hot garbage. It doesn’t properly reason, it can’t design architecture, it can’t write code that is comprehensible to other programmers, and treating it as a “black box to be manipulated by AI” just leads to dead ends that can’t be escaped, terrible decisions that will take huge amounts of expert coding time to undo, subtle bugs that AI can’t fix and are super hard to spot, and often you can’t understand their code enough to fix them, and security nightmares.
Testing is insufficient for good code. Humans write code in a way that is designed for general correctness. AI does not, at least not yet.
I do think these problems can be solved. I think we probably need automated reasoning systems, or else vastly improved LLMs that border on automated reasoning much like humans do. Could be a year. Could be a decade. But right now these tools don’t work well. Great for vibe coding, prototyping, analysis, review, bouncing ideas.
Please tell me which one of the headings is not about increased usage o LLMs and derived tools and is about some improvement in the axes of reliability or or any kind of usefulness.
> The coding agent heading. Claude Code and tools like it represent a huge improvement in what you can usefully get done with LLMs.
Does it? It's all prompt manipulation. Shell script are powerful yes, but not really huge improvement over having a shell (REPL interface) to the system. And even then a lot of programs just use syscalls or wrapper libraries.
> can try the code, see that it doesn't work and fix the problem.
If you mean correct often enough that you can expect it to be a productive assistant that helps solve all sorts of problems faster than you could solve them without it, and which makes mistakes infrequently enough that you waste less time fixing them than you would doing everything by yourself then yes, it's plenty reliable enough now.
I know it seems like forever ago, but claude code only came out in 2025.
Its very difficult to argue the point that claude code:
1) was a paradigm shift in terms of functionality, despite, to be fair, at best, incremental improvements in the underlying models.
2) The results are an order of magnitude, I estimate, better in terms of output.
I think its very fair to distill “AI progress 2025” to: you can get better results (up to a point; better than raw output anyway; scaling to multiple agents has not worked) without better models with clever tools and loops. (…and video/image slop infests everything :p).
Did more software ship in 2025 than in 2024? I'm still looking for some actual indication of output here. I get that people feel more productive but the actual metrics don't seem to agree.
I'm still waiting for the Linux drivers to be written because of all the 20x improvements that AI hypers are touting. I would even settle for Apple M3 and M4 computers to be supported by Asahi.
Whenever someone tells me that AI is worthless, does nothing, scam/slop etc, I ask them about their own AI usage, and their general knowledge about what's going on.
Invariably they've never used AI, or at most very rarely. (If they used AI beyond that, this would be admission that it was useful at some level).
Therefore it's reasonable to assume that you are in that boat. Now that might not be true in your case, who knows, but it's definitely true on average.
This comment is legitimately hilarious to me. I thought it was satire at first. The list of what has happened in this field in the last twelve months is staggering to me, while you write it off as essentially nothing.
Different strokes, but I’m getting so much more done and mostly enjoying it. Can’t wait to see what 2026 holds!
People who dislike LLMs are generally insistent that they're useless for everything and have infinitely negative value, regardless of facts they're presented with.
Anyone that believes that they are completely useless is just as deluded as anyone that believes they're going to bring an AGI utopia next week.
The big labs are (mostly) investing a lot of resources into reducing the chance their models will trigger self-harm and AI psychosis and suchlike. See the GPT-4o retirement (and resulting backlash) for an example of that.
But the number of users is exploding too. If they make things 5x less likely to happen but sign up 10x more people it won't be good on that front.
Sure -- but that's fair game in engineering. I work on cars. If we kill people with safety faults I expect it to make more headlines than all the fun roadtrips.
What I find interesting with chat bots is that they're "web apps" so to speak, but with safety engineering aspects that type of developer is typically not exposed to or familiar with.
One of the tough problems here is privacy. AI labs really don't want to be in the habit of actively monitoring people's conversations with their bots, but they also need to prevent bad situations from arising and getting worse.
Until AI labs have the equivalent of an SLA for giving accurate and helpful responses it don't get better. They've not even able to measure if the agents work correctly and consistently.
I'm curious how all of the progress will be seen if it does indeed result in mass unemployment (but not eradication) of professional software engineers.
I nearly added a section about that. I wanted to contrast the thing where many companies are reducing junior engineering hires with the thing where Cloudflare and Shopify are hiring 1,000+ interns. I ran out of time and hadn't figured out a good way to frame it though so I dropped it.
My prediction: If we can successfully get rid of most software engineers, we can get rid of most knowledge work. Given the state of robotics, manual labor is likely to outlive intellectual labor.
"Given the state of robotics" reminds me a lot of what was said about llms and image/video models over the past 3 years. Considering how much llms improved, how long can robotics be in this state?
I have to think 3 years from now we will be having the same conversation about robots doing real physical labor.
"This is the worst they will ever be" feels more apt.
Crypto bros in hindsight were so much less dangerous than AI bros. At least they weren't trying to construct data centers in rural America or prop up artificial stocks like $NVDA.
Nothing about the severe impact on the environment, and the hand waviness about water usage hurt to read. The referenced post was missing every single point about the issue by making it global instead of local. And as if data center buildouts are properly planned and dimensioned for existing infrastructure…
Add to this that all the hardware is already old and the amount of waste we’re producing right now is mind boggling, and for what, fun tools for the use of one?
I don’t live in the US, but the amount of tax money being siphoned to a few tech bros should have heads rolling and I really don’t want to see it happening in Europe.
But I guess we got a new version number on a few models and some blown up benchmarks so that’s good, oh and of course the svg images we will never use for anything.
I never quite got what was so "hot" about it. There seems to be an entire parallel ecosystem of corporates that are just begging to turn AI into PowerPoint slides so that they can mould it into a shape that's familiar.
> The year of YOLO and the Normalization of Deviance #
On this including AI agents deleting home folders, I was able to run agents in Firejail by isolating vscode (Most of my agents are vscode based ones, like Kilo Code).
Took a bit of tweaking, vscode crashing a bunch of times with not being able to read its config files, but I got there in the end. Now it can only write to my projects folder. All of my projects are backed up in git.
You’re absolutely right! You astutely observed that 2025 was a year with many LLMs and this was a selection of waypoints, summarized in a helpful timeline.
That’s what most non-tech-person’s year in LLMs looked like.
Hopefully 2026 will be the year where companies realize that implementing intrusive chatbots can’t make better ::waving hands:: ya know… UX or whatever.
For some reason, they think its helpful to distractingly pop up chat windows on their site because their customers need textual kindergarten handholding to … I don’t know… find the ideal pocket comb for their unique pocket/hair situation, or had an unlikely question about that aerosol pan release spray that a chatbot could actually answer. Well, my dog also thinks she’s helping me by attacking the vacuum when I’m trying to clean. Both ideas are equally valid.
And spending a bazillion dollars implementing it doesn’t mean your customers won’t hate it. And forcing your customers into pathways they hate because of your sunk costs mindset means it will never stop costing you more money than it makes.
I just hope companies start being honest with themselves about whether or not these things are good, bad, or absolutely abysmal for the customer experience and cut their losses when it makes sense.
They need to be intrusive and shoved in your face. This way, they can say they have a lot of people using them, which is a good and useful metric.
> For some reason, they think its helpful to distractingly pop up chat windows on their site...
Companies have been doing this "live support" nonsense far longer than LLMs have been popular.
Pretty much a whole year of nothing really. Just coming with a bunch of abstraction and ideas trying to solve an unsolvable problem. Getting reliable results from an unreliable process while assuming the process is reliable.
At least when herding cats, you can be sure that if the cats are hungry, they will try to get where the food is.
This is extremely dismissive. Claude Code helps me make a majority of changes to our codebase now, particularly small ones, and is an insane efficiency boost. You may not have the same experience for one reason or another, but plenty of devs do, so "nothing happened" is absolutely wrong.
2024 was a lot of talk, a lot of "AI could hypothetically do this and that". 2025 was the year where it genuinely started to enter people's workflows. Not everything we've been told would happen has happened (I still make my own presentations and write my own emails) but coding agents certainly have!
Did you ship more in 2025 than in 2024?
I definitely did.
I’m not sure how to tell you how obvious it is you haven’t actually used these tools.
Why do people assume negative critique is ignorance?
People denied that bicycles could possibly balance even as others happily pedaled by. This is the same thing.
people also said that selling jpegs of monkeys for millions of dollars was a pump and dump scam, and would collapse
they were right
Bicycles don't balance, the human on the bicycle is the one doing the balancing.
Yes, that is the analogy I am making. People argued that bicycles (a tool for humans to use) could not possibly work - even as people were successfully using them.
People use drugs as well but I'm not sure I'd call that successful use of chemical compounds without further context. There are many analogies one can apply here that would be equally valid.
Yikes.
It’s possible this is correct.
It’s also possible that people more experienced, knowledgable and skilled than you can see fundamental flaws in using LLMs for software engineering that you cannot. I am not including myself in that category.
I’m personally honestly undecided. I’ve been coding for over 30 years and know something like 25 languages. I’ve taught programming to postgrad level, and built prototype AI systems that foreshadowed LLMs, I’ve written everything from embedded systems to enterprise, web, mainframes, real time, physics simulation and research software. I would consider myself an 7/10 or 8/10 coder.
A lot of folks I know are better coders. To put my experience into context: one guy in my year at uni wrote one of the world’s most famous crypto systems; another wrote large portions of some of the most successful games of the last few decades. So I’ve grown up surrounded by geniuses, basically, and whilst I’ve been lectured by true greats I’m humble enough to recognise I don’t bleed code like they do. I’m just a dabbler. But it irks me that a lot of folks using AI profess it’s the future but don’t really know anything about coding compared to these folks. Not to be a Luddite - they are the first people to adopt new languages and techniques, but they also are super sceptical about anything that smells remotely like bullshit.
One of the most wise insights in coding is the aphorism“beware the enthusiasm of the recently converted.” And I see that so much with AI. I’ve seen it with compilers, with IDEs, paradigms, and languages.
I’ve been experimenting a lot with AI, and I’ve found it fantastic for comprehending poor code written by others. I’ve also found it great for bouncing ideas. And the code it writes, beyond boiler plate, is hot garbage. It doesn’t properly reason, it can’t design architecture, it can’t write code that is comprehensible to other programmers, and treating it as a “black box to be manipulated by AI” just leads to dead ends that can’t be escaped, terrible decisions that will take huge amounts of expert coding time to undo, subtle bugs that AI can’t fix and are super hard to spot, and often you can’t understand their code enough to fix them, and security nightmares.
Testing is insufficient for good code. Humans write code in a way that is designed for general correctness. AI does not, at least not yet.
I do think these problems can be solved. I think we probably need automated reasoning systems, or else vastly improved LLMs that border on automated reasoning much like humans do. Could be a year. Could be a decade. But right now these tools don’t work well. Great for vibe coding, prototyping, analysis, review, bouncing ideas.
People did?
Please tell me which one of the headings is not about increased usage o LLMs and derived tools and is about some improvement in the axes of reliability or or any kind of usefulness.
Here is the changelog for OpenBSD 7.8:
https://www.openbsd.org/78.html
There's nothing here that says: We make it easier to use it more of it. It's about using it better and fixing underlying problems.
The coding agent heading. Claude Code and tools like it represent a huge improvement in what you can usefully get done with LLMs.
Mistakes and hallucinations matter a whole lot less if a reasoning LLM can try the code, see that it doesn't work and fix the problem.
> The coding agent heading. Claude Code and tools like it represent a huge improvement in what you can usefully get done with LLMs.
Does it? It's all prompt manipulation. Shell script are powerful yes, but not really huge improvement over having a shell (REPL interface) to the system. And even then a lot of programs just use syscalls or wrapper libraries.
> can try the code, see that it doesn't work and fix the problem.
Can you really say that does happens reliably?
Depends on what you mean by "reliably".
If you mean 100% correct all of the time then no.
If you mean correct often enough that you can expect it to be a productive assistant that helps solve all sorts of problems faster than you could solve them without it, and which makes mistakes infrequently enough that you waste less time fixing them than you would doing everything by yourself then yes, it's plenty reliable enough now.
If it actually does that without an argument. I can't believe I have to say that about a computer program
I know it seems like forever ago, but claude code only came out in 2025.
Its very difficult to argue the point that claude code:
1) was a paradigm shift in terms of functionality, despite, to be fair, at best, incremental improvements in the underlying models.
2) The results are an order of magnitude, I estimate, better in terms of output.
I think its very fair to distill “AI progress 2025” to: you can get better results (up to a point; better than raw output anyway; scaling to multiple agents has not worked) without better models with clever tools and loops. (…and video/image slop infests everything :p).
Did more software ship in 2025 than in 2024? I'm still looking for some actual indication of output here. I get that people feel more productive but the actual metrics don't seem to agree.
I'm still waiting for the Linux drivers to be written because of all the 20x improvements that AI hypers are touting. I would even settle for Apple M3 and M4 computers to be supported by Asahi.
Whenever someone tells me that AI is worthless, does nothing, scam/slop etc, I ask them about their own AI usage, and their general knowledge about what's going on.
Invariably they've never used AI, or at most very rarely. (If they used AI beyond that, this would be admission that it was useful at some level).
Therefore it's reasonable to assume that you are in that boat. Now that might not be true in your case, who knows, but it's definitely true on average.
because your "negative critique" is just idiotic and wrong
This comment is legitimately hilarious to me. I thought it was satire at first. The list of what has happened in this field in the last twelve months is staggering to me, while you write it off as essentially nothing.
Different strokes, but I’m getting so much more done and mostly enjoying it. Can’t wait to see what 2026 holds!
People who dislike LLMs are generally insistent that they're useless for everything and have infinitely negative value, regardless of facts they're presented with.
Anyone that believes that they are completely useless is just as deluded as anyone that believes they're going to bring an AGI utopia next week.
Not in this review: Also the record year in intelligent systems aiding in and prompting human users into fatal self-harm.
Will 2026 fare better?
The people working on this stuff have convinced themselves they're on a religious quest so it's not going to get better: https://x.com/RobertFreundLaw/status/2006111090539687956
I really hope so.
The big labs are (mostly) investing a lot of resources into reducing the chance their models will trigger self-harm and AI psychosis and suchlike. See the GPT-4o retirement (and resulting backlash) for an example of that.
But the number of users is exploding too. If they make things 5x less likely to happen but sign up 10x more people it won't be good on that front.
Also essential self-fulfilment.
But that one doesn't make headlines ;)
Sure -- but that's fair game in engineering. I work on cars. If we kill people with safety faults I expect it to make more headlines than all the fun roadtrips.
What I find interesting with chat bots is that they're "web apps" so to speak, but with safety engineering aspects that type of developer is typically not exposed to or familiar with.
One of the tough problems here is privacy. AI labs really don't want to be in the habit of actively monitoring people's conversations with their bots, but they also need to prevent bad situations from arising and getting worse.
Until AI labs have the equivalent of an SLA for giving accurate and helpful responses it don't get better. They've not even able to measure if the agents work correctly and consistently.
I'm curious how all of the progress will be seen if it does indeed result in mass unemployment (but not eradication) of professional software engineers.
I nearly added a section about that. I wanted to contrast the thing where many companies are reducing junior engineering hires with the thing where Cloudflare and Shopify are hiring 1,000+ interns. I ran out of time and hadn't figured out a good way to frame it though so I dropped it.
My prediction: If we can successfully get rid of most software engineers, we can get rid of most knowledge work. Given the state of robotics, manual labor is likely to outlive intellectual labor.
"Given the state of robotics" reminds me a lot of what was said about llms and image/video models over the past 3 years. Considering how much llms improved, how long can robotics be in this state?
I have to think 3 years from now we will be having the same conversation about robots doing real physical labor.
"This is the worst they will ever be" feels more apt.
Remember, back in the day, when a year of progress was like, oh, they voted to add some syntactic sugar to Java...
> they voted to add some syntactic sugar to Java...
I remember when we just wanted to rewrite everything in Rust.
Those were the simpler times, when crypto bros seemed like the worst venture capitalism could conjure.
Crypto bros in hindsight were so much less dangerous than AI bros. At least they weren't trying to construct data centers in rural America or prop up artificial stocks like $NVDA.
forgot to mention the first murder-suicide instigated by chatgpt
Great summary of the year in LLMs. Is there a predictions (for 2026) blogpost as well?
Given how badly my 2025 predictions aged I'm probably going to sit that one out! https://simonwillison.net/2025/Jan/10/ai-predictions/
Nothing about the severe impact on the environment, and the hand waviness about water usage hurt to read. The referenced post was missing every single point about the issue by making it global instead of local. And as if data center buildouts are properly planned and dimensioned for existing infrastructure…
Add to this that all the hardware is already old and the amount of waste we’re producing right now is mind boggling, and for what, fun tools for the use of one?
I don’t live in the US, but the amount of tax money being siphoned to a few tech bros should have heads rolling and I really don’t want to see it happening in Europe.
But I guess we got a new version number on a few models and some blown up benchmarks so that’s good, oh and of course the svg images we will never use for anything.
"Nothing about the severe impact on the environment"
I literally said:
"AI data centers continue to burn vast amounts of energy and the arms race to build them continues to accelerate in a way that feels unsustainable."
AND I linked to my coverage from last year, which is still true today (hence why I felt no need to update it): https://simonwillison.net/2024/Dec/31/llms-in-2024/#the-envi...
These are excellent every year, thank you for all the wonderful work you do.
Same here. Simon is one of the main reasons I’ve been able to (sort of) keep up with developments in AI.
I look forward to learning from his blog posts and HN comments in the year ahead, too.
> The (only?) year of MCP
I like to believe, but MCP is quickly turning into an enterprise thing so I think it will stick around for good.
I think it will stick around, but I don't think it will have another year where it's the hot thing it was back in January through May.
I never quite got what was so "hot" about it. There seems to be an entire parallel ecosystem of corporates that are just begging to turn AI into PowerPoint slides so that they can mould it into a shape that's familiar.
What an amazing progress in just short time. The future is bright! Happy New Year y'all!
> Vendor-independent options include GitHub Copilot CLI, Amp, OpenHands CLI, and Pi
...and the best of them all, OpenCode[1] :)
[1]: https://opencode.ai
I don't know why you're downloaded, OpenCode is by far the best.
Good call, I'll add that. I think I mentally scrambled it with OpenHands.
Thanks for adding pi to it though :)
How did I miss this until now! Thank you for sharing.
> The year of YOLO and the Normalization of Deviance #
On this including AI agents deleting home folders, I was able to run agents in Firejail by isolating vscode (Most of my agents are vscode based ones, like Kilo Code).
I wrote a little guide on how I did it https://softwareengineeringstandard.com/2025/12/15/ai-agents...
Took a bit of tweaking, vscode crashing a bunch of times with not being able to read its config files, but I got there in the end. Now it can only write to my projects folder. All of my projects are backed up in git.
2025: The Year in LLMs
I will never stop treating hallucinations as inventions. I dare you to stop me. i double dog dare y