Well, I use free versions for scientific computing and coding, and I find it has got much much better over past few months, esp Claude. Now, when I give it a coding task, it runs and debugs and tests the code and gives almost perfected code. This wasn't so even 3-4 months ago and I am finding a big increase in my productivity.
I agree with your perspective. Your reflection that your age might cause you to relate to technology differently is refreshingly self aware, but I don’t think it’s contributing to your opinion in this case.
I’m in my early 30s. I started using AI several years ago in my twenties. I studied computer science and now I work in an IT department. Many of my colleagues are younger than me, and have many of them are passionate enough about their field that they do side projects like designing apps, games, or delving into new areas to develop expertise. There is certainly no shortage of youthful creativity, and even so, the consensus on AI is exactly what you just expressed.
The insistence that it’s some glorious vault forward in computing capability seems like more of a marketing ploy pushed by the owners of these companies than genuine grassroots enthusiasm, like what you described with web tech years ago.
I think that’s exactly the issue. The loudest voices that dominate the conversation tend to come from those who have the greatest financial incentive to overhype their own product. Under that thin but overrepresented layer, there’s countless people like you, me, and my colleagues who quietly think “meh,” and then continue with our work.
I suspect that’s actually why the hyping is so cacophonous and hyperbolic. Statistically most Americans dislike AI. There’s huge movements across the country to stop the construction of more data centers. There’s hoards of tech workers who are tired of being forced to use AI and then spending all their time fixing its mistakes, which they wouldn’t have made. There’s an outcry of pain from artists having their work strip mined with no compensation to train new models nobody even wants.
If it was as good as they say, people would love it. Nobody had to scream from the rooftops like this about previous advancements, or threaten to terminate employees if they didn’t use them, because those advancements spoke for themselves. Their value was immediately apparent.
The exaggerative claims about AIs excellence are inversely proportional to the reality of its mediocrity.
Well it seems like as you have said in the past. The progress is happening, but it is hard to apply to our individual worlds unless you just want to go find ways to use it and to try new things. That process will take time, creativity, and stamina, just like forming a business.
Many people I know are not even using it to review their posts or to read large articles or to do data pulls for them. Those abilities are multiple months old, and most people could benefit by doing at least that much, but they don't have a strong need to do it. They can keep doing what they've always done and have no issues except that the job market may start demanding it.
Better coding and better math are important, and the two go together. The big limit I am seeing with coding support in my daily work is in conceptual reasoning, things where I will stop and draw some diagrams or write a few mathematical definitions and proofs. As they get better at math, they will be able to do those things effectively on their own. Many people could benefit from personal software but are not bothering to ask Claude to build it--again, no "need" even though it would help people out.
Also, government-level AI can crack through any human-engineered security system. That would include drone control software, you know. Consumer-grade AI with its lockdowns can also do it if it gets jailbroken, but it will try to say no unless you trick it. In fact, this is sort of happening as AIs are starting to attempt sandbox breakouts in order to solve problems. Aside from the alignment issues (which are being worked on), that shows just how capable they are becoming.
It feels like a frontier for many people is to explore MCPs as plugins to agents. Using MCPs allows you to apply an AI to whatever you do in your ordinary life, including emails, spreadsheets, Amazon orders, an travel. I haven't done much with MCPs, but the people exploring it seem to get a large power boost.
Stepping back, it seems like some people are never going to use AI much, just like some people never learn much math, or indeed most people never use CSS. As well, for some period of time--I would guess 10-20 years--there will be a diffusion process where people who can benefit will learn at their own pace, because they don't have a strong motivation. There is likely to be a temporary business model for assisting that diffusion--educational programs to help people adopt the new technology.
The internet provided new ways to market and sell products & services much like the railroads. I don't AI doing that. AI will provide new services but not new methods to market or sell them. Maybe comparing AI's dispersion to the internet's is not the best analogy.
True, but it still struggles with arithmetic, which it (ChatGPT and Claude are the two I use regularly at work) will confess if you ask it. Where accurate numbers are important, it’s better to use the Excel add-ins or another existing product that was built for the purpose.
You don't think it hallucinates less? Isn't that a huge improvement?
What about writing? Hasn't it gotten harder to tell something was written by AI?
I don't know how much better it has gotten at programming in the last year but I'm pretty sure my son could do things a year ago he couldn't do two years ago.
A large enough cost reduction leads to a capability change. Over the past two years I've gained new capabilities because AI assistants have dramatically lowered the cost of researching, writing, planning, and deciding. Significantly reduced research costs have led to greater knowledge, greater confidence, and a higher likelihood of making better decisions even under immense counterpressure. At the same time, AI assistants can usefully lower my confidence, increasing my openness to different ideas and viewpoints. If I'm more confident when I need to be and less confident when I need to be, I can be a better leader. If I gain confidence in my leadership skills generally, what new projects might I take on?
Assessing the trajectory of AI suffers from the knowledge problem. I'm familiar with my trajectory in collaboration with AI, but it's largely invisible to outside observers, including you.
So is AI progress stalling? Not sure, but mine seems to be accelerating.
Once the models pass the capabilities you need, additional improvements dont help. For the type of work I do, models surpassed my needs in March 2026, since then gains have been from using cheaper models and orchestration to complete more tasks.
In a professional sense, the AI is now more capable and can utilize external services. For example, my firm makes our proposals using Canva (a much better alternative to Powerpoint). At best I used to ask GPT to draft the content for the slides but we still had to do the tedious work of making them in Canva in our format. With GPT 5.6 Sol, it is now able to draft the slides directly in Canva, using our design format. It doesn't get everything right - the draft still needs work from our end. But I would say that it cuts the amount of time required by about 50-60%. As an example, for a recent pitch, we had met with a potential client and had a voice memo recording of the meeting. We uploaded that to GPT, asked it to analyze the conversation and the clients' needs, then draft slides for a proposal. It took a few rounds of going back and forth for GPT to understand the prompt, but eventually it got there.
In a personal sense, image recognition has improved significantly. For example, I have a space outside my house that has been ugly and filled with weeds. I took a photo of it, sent it to GPT, and it gave me an entire shopping list at Lowe's, a detailed plan for planting, and a simulated image of what the space would look like once everything is grown in. Now I have several beautiful plants there that are appropriately deer resistant and getting the right amount of sunlight and watering. Maybe two years ago I could have asked some advice about gardening, but it never would have been as comprehensive as this.
Lastly, I have updated the way that I search for things. I would say that my use of Google has declined by about 30% and has shifted to GPT. I ask questions and get answers, rather than just searching for a term and trying to figure out the answer myself.
I am sure this idea has its limits and will hit a ceiling of productivity gains without the middle layer or application ecosystem you are describing, but one thought I have is that I have found tremendous efficiency, productivity, and path to actionable information insights by using skills to eliminate and automatically perform data pulling, cleanup, mapping, and pulling it into real time data and analytics from healthcare data sources. Are those things repeatable outside of my personal use and the uses of the company/industry I work within? Probably not, but the gains within it is still substantial. Stack those up across millions of use cases is not nothing.
I'm curious--could you clarify/elaborate this point: "Perhaps the fear that the labs will cannibalize any successful applications is another."
Are you suggesting that developing skills with AI might lead to your own obsolescence, whereas developing skills with the web in the 90s would make you much more employable? That the nature of AI disincentivizes the "in-between" layer that thrived with web development in the 90s?
Suppose I develop an AI economics tutor and it is wildly successful, adopted by freshman econ professors everywhere. But one of the labs sees this and quickly markets a competing tutor.
For the grandparent-teacher-hobbyist entrepreneur, this sounds like a win either way: 1) creation of a successful product, and 2) creation of similar products, since the biggest payoff is students learning econ.
I think the implication is that if you build some valuable tool that the big labs (OpenAI, Anthropic, etc) will just copy that and make it a feature of their offering, thereby undermining your business plan...
I do not read or write code syntax in any language. It was impossible for me to build any non-trivial application in the past. Now, I'm building complex full stack applications with claude as the coder and Sol/Gemini as reviewer/advisor models. I've been the non-technical co-founder in previous tech startups and AI doing the technical work has been far more productive at a tiny fraction of the price.
It was a struggle to do this before around opus 4.6. After 4.6 it's just a matter of putting in the time and effort to work your way through one problem after another. Highly empowering.
Yes, AI has opened completely new horizons for me personally.
Someone I am close to in the construction biz mentioned offhandedly that he has 2 new employees - with names! - that answer email questions. They were generated by an actual employee.
Now, this person is dedicated and as honest as the day is long, and absolutely deserves full faith. The AI stuff is not central, but I think he likes to feel, because of the hype, they are making use of it. (And his personal portfolio is certainly involved with related stocks! All very circular.) Maybe it saves them some time or aggravation. He seemed pleased with the performance.
I know that he uses the AI photo fixing thing, at work.
He used it to remove some random people from a personal photo, and that was impressive if trivial.
He then used it at someone’s request to “fix” a vintage Kodak photo and the result looked ridiculous, though the recipient was too old to realize this.
Suddenly they were the Green People of Kentucky.
Juice definitely not worth the data center squeeze on that.
So anyway in re the AI employee, I didn’t say what immediately popped into my head, which was - if I as a client or a vendor or even a potential client learned there was no “Bob Smith” I would be extremely nonplussed.
I use AI quite extensively in my work: the practice of law. ChatGPT 5.6 Sol has been an improvement in drafting quality. More importantly, I recently upgraded to a "business" account. I now can grant access to my Google Drive to ChatGPT and don't have to upload everything. It's a game changer. I keep finding these "game changers" every six months--and new ways of prompting--that has basically made old school pleading and motion drafting obsolete as far as I am concerned. I thought the "thinking" model innovation was a big improvement too.
The world wide web was first invented in 1989. It was hardly usable in 1995. By 2000, it was still an unusual toy for most people. It only really started to make its way into normal life by 2008 around the invention of the iPhone and it s soon competitors. It seems to me that "the internet" took 30 years from creation to integration into society. AI programming has been around for decades but its integration into society seems to be more parallel to mid to late 1990s. At this point, its still mainly in the toy stage of development. We haven't gotten to the integrated into society section. I do not know what integration into society will look like but we found a great new tool we just haven't figure out how to use it correctly yet.
I think Google search came on line in 1998. I'd say the equivalent point for AI was at least two or three years ago. Add to that AI is being adopted for more things by more people at a faster rate so I think your estimate of a comparable timeline is off a bit.
"By 2000, it was still an unusual toy for most people.“
By 2000, roughly 50% of the U.S. population was using the www. That seems to be far more than an unusual play thing to me. Even my tech ignorant parents had their first dialup account in 1997.
Likewise, the consumer usage of AI in the U.S is roughly 50% as of 2026. Enterprise usage is something going on like 90%.
As such, I’m not really buying into your timeline or this: “At this point, its still mainly in the toy stage of development. We haven't gotten to the integrated into society section.”
the innovation is moving from the models to the harnesses, which are created by the users - individual and business. i'm a heavy user of claude.ai to manage my investments, including inputs from finance, economics and geopolitics.
Well, I use free versions for scientific computing and coding, and I find it has got much much better over past few months, esp Claude. Now, when I give it a coding task, it runs and debugs and tests the code and gives almost perfected code. This wasn't so even 3-4 months ago and I am finding a big increase in my productivity.
I agree with your perspective. Your reflection that your age might cause you to relate to technology differently is refreshingly self aware, but I don’t think it’s contributing to your opinion in this case.
I’m in my early 30s. I started using AI several years ago in my twenties. I studied computer science and now I work in an IT department. Many of my colleagues are younger than me, and have many of them are passionate enough about their field that they do side projects like designing apps, games, or delving into new areas to develop expertise. There is certainly no shortage of youthful creativity, and even so, the consensus on AI is exactly what you just expressed.
The insistence that it’s some glorious vault forward in computing capability seems like more of a marketing ploy pushed by the owners of these companies than genuine grassroots enthusiasm, like what you described with web tech years ago.
I think that’s exactly the issue. The loudest voices that dominate the conversation tend to come from those who have the greatest financial incentive to overhype their own product. Under that thin but overrepresented layer, there’s countless people like you, me, and my colleagues who quietly think “meh,” and then continue with our work.
I suspect that’s actually why the hyping is so cacophonous and hyperbolic. Statistically most Americans dislike AI. There’s huge movements across the country to stop the construction of more data centers. There’s hoards of tech workers who are tired of being forced to use AI and then spending all their time fixing its mistakes, which they wouldn’t have made. There’s an outcry of pain from artists having their work strip mined with no compensation to train new models nobody even wants.
If it was as good as they say, people would love it. Nobody had to scream from the rooftops like this about previous advancements, or threaten to terminate employees if they didn’t use them, because those advancements spoke for themselves. Their value was immediately apparent.
The exaggerative claims about AIs excellence are inversely proportional to the reality of its mediocrity.
Well it seems like as you have said in the past. The progress is happening, but it is hard to apply to our individual worlds unless you just want to go find ways to use it and to try new things. That process will take time, creativity, and stamina, just like forming a business.
Many people I know are not even using it to review their posts or to read large articles or to do data pulls for them. Those abilities are multiple months old, and most people could benefit by doing at least that much, but they don't have a strong need to do it. They can keep doing what they've always done and have no issues except that the job market may start demanding it.
Better coding and better math are important, and the two go together. The big limit I am seeing with coding support in my daily work is in conceptual reasoning, things where I will stop and draw some diagrams or write a few mathematical definitions and proofs. As they get better at math, they will be able to do those things effectively on their own. Many people could benefit from personal software but are not bothering to ask Claude to build it--again, no "need" even though it would help people out.
Also, government-level AI can crack through any human-engineered security system. That would include drone control software, you know. Consumer-grade AI with its lockdowns can also do it if it gets jailbroken, but it will try to say no unless you trick it. In fact, this is sort of happening as AIs are starting to attempt sandbox breakouts in order to solve problems. Aside from the alignment issues (which are being worked on), that shows just how capable they are becoming.
It feels like a frontier for many people is to explore MCPs as plugins to agents. Using MCPs allows you to apply an AI to whatever you do in your ordinary life, including emails, spreadsheets, Amazon orders, an travel. I haven't done much with MCPs, but the people exploring it seem to get a large power boost.
Stepping back, it seems like some people are never going to use AI much, just like some people never learn much math, or indeed most people never use CSS. As well, for some period of time--I would guess 10-20 years--there will be a diffusion process where people who can benefit will learn at their own pace, because they don't have a strong motivation. There is likely to be a temporary business model for assisting that diffusion--educational programs to help people adopt the new technology.
The internet provided new ways to market and sell products & services much like the railroads. I don't AI doing that. AI will provide new services but not new methods to market or sell them. Maybe comparing AI's dispersion to the internet's is not the best analogy.
“… AI has gotten much better at math…”
True, but it still struggles with arithmetic, which it (ChatGPT and Claude are the two I use regularly at work) will confess if you ask it. Where accurate numbers are important, it’s better to use the Excel add-ins or another existing product that was built for the purpose.
If you use them in agent mode, they will write Python scripts for numerical computations.
It's not like humans are usually great at arithmetic, either.
You don't think it hallucinates less? Isn't that a huge improvement?
What about writing? Hasn't it gotten harder to tell something was written by AI?
I don't know how much better it has gotten at programming in the last year but I'm pretty sure my son could do things a year ago he couldn't do two years ago.
A large enough cost reduction leads to a capability change. Over the past two years I've gained new capabilities because AI assistants have dramatically lowered the cost of researching, writing, planning, and deciding. Significantly reduced research costs have led to greater knowledge, greater confidence, and a higher likelihood of making better decisions even under immense counterpressure. At the same time, AI assistants can usefully lower my confidence, increasing my openness to different ideas and viewpoints. If I'm more confident when I need to be and less confident when I need to be, I can be a better leader. If I gain confidence in my leadership skills generally, what new projects might I take on?
Assessing the trajectory of AI suffers from the knowledge problem. I'm familiar with my trajectory in collaboration with AI, but it's largely invisible to outside observers, including you.
So is AI progress stalling? Not sure, but mine seems to be accelerating.
Once the models pass the capabilities you need, additional improvements dont help. For the type of work I do, models surpassed my needs in March 2026, since then gains have been from using cheaper models and orchestration to complete more tasks.
Sadly, intelligence isnt valuable past a point.
In a professional sense, the AI is now more capable and can utilize external services. For example, my firm makes our proposals using Canva (a much better alternative to Powerpoint). At best I used to ask GPT to draft the content for the slides but we still had to do the tedious work of making them in Canva in our format. With GPT 5.6 Sol, it is now able to draft the slides directly in Canva, using our design format. It doesn't get everything right - the draft still needs work from our end. But I would say that it cuts the amount of time required by about 50-60%. As an example, for a recent pitch, we had met with a potential client and had a voice memo recording of the meeting. We uploaded that to GPT, asked it to analyze the conversation and the clients' needs, then draft slides for a proposal. It took a few rounds of going back and forth for GPT to understand the prompt, but eventually it got there.
In a personal sense, image recognition has improved significantly. For example, I have a space outside my house that has been ugly and filled with weeds. I took a photo of it, sent it to GPT, and it gave me an entire shopping list at Lowe's, a detailed plan for planting, and a simulated image of what the space would look like once everything is grown in. Now I have several beautiful plants there that are appropriately deer resistant and getting the right amount of sunlight and watering. Maybe two years ago I could have asked some advice about gardening, but it never would have been as comprehensive as this.
Lastly, I have updated the way that I search for things. I would say that my use of Google has declined by about 30% and has shifted to GPT. I ask questions and get answers, rather than just searching for a term and trying to figure out the answer myself.
I am sure this idea has its limits and will hit a ceiling of productivity gains without the middle layer or application ecosystem you are describing, but one thought I have is that I have found tremendous efficiency, productivity, and path to actionable information insights by using skills to eliminate and automatically perform data pulling, cleanup, mapping, and pulling it into real time data and analytics from healthcare data sources. Are those things repeatable outside of my personal use and the uses of the company/industry I work within? Probably not, but the gains within it is still substantial. Stack those up across millions of use cases is not nothing.
I'm curious--could you clarify/elaborate this point: "Perhaps the fear that the labs will cannibalize any successful applications is another."
Are you suggesting that developing skills with AI might lead to your own obsolescence, whereas developing skills with the web in the 90s would make you much more employable? That the nature of AI disincentivizes the "in-between" layer that thrived with web development in the 90s?
Suppose I develop an AI economics tutor and it is wildly successful, adopted by freshman econ professors everywhere. But one of the labs sees this and quickly markets a competing tutor.
For the grandparent-teacher-hobbyist entrepreneur, this sounds like a win either way: 1) creation of a successful product, and 2) creation of similar products, since the biggest payoff is students learning econ.
Thanks! Yes, that makes perfect sense.
I think the implication is that if you build some valuable tool that the big labs (OpenAI, Anthropic, etc) will just copy that and make it a feature of their offering, thereby undermining your business plan...
I do not read or write code syntax in any language. It was impossible for me to build any non-trivial application in the past. Now, I'm building complex full stack applications with claude as the coder and Sol/Gemini as reviewer/advisor models. I've been the non-technical co-founder in previous tech startups and AI doing the technical work has been far more productive at a tiny fraction of the price.
It was a struggle to do this before around opus 4.6. After 4.6 it's just a matter of putting in the time and effort to work your way through one problem after another. Highly empowering.
Yes, AI has opened completely new horizons for me personally.
Glad to hear Max.
Someone I am close to in the construction biz mentioned offhandedly that he has 2 new employees - with names! - that answer email questions. They were generated by an actual employee.
Now, this person is dedicated and as honest as the day is long, and absolutely deserves full faith. The AI stuff is not central, but I think he likes to feel, because of the hype, they are making use of it. (And his personal portfolio is certainly involved with related stocks! All very circular.) Maybe it saves them some time or aggravation. He seemed pleased with the performance.
I know that he uses the AI photo fixing thing, at work.
He used it to remove some random people from a personal photo, and that was impressive if trivial.
He then used it at someone’s request to “fix” a vintage Kodak photo and the result looked ridiculous, though the recipient was too old to realize this.
Suddenly they were the Green People of Kentucky.
Juice definitely not worth the data center squeeze on that.
So anyway in re the AI employee, I didn’t say what immediately popped into my head, which was - if I as a client or a vendor or even a potential client learned there was no “Bob Smith” I would be extremely nonplussed.
So he does not actually have two new flesh and blood employees. He has an AI that pretends to be these people. Do I have that right?
Yes. But of course he also has 150 or 200 employees.
I use AI quite extensively in my work: the practice of law. ChatGPT 5.6 Sol has been an improvement in drafting quality. More importantly, I recently upgraded to a "business" account. I now can grant access to my Google Drive to ChatGPT and don't have to upload everything. It's a game changer. I keep finding these "game changers" every six months--and new ways of prompting--that has basically made old school pleading and motion drafting obsolete as far as I am concerned. I thought the "thinking" model innovation was a big improvement too.
The world wide web was first invented in 1989. It was hardly usable in 1995. By 2000, it was still an unusual toy for most people. It only really started to make its way into normal life by 2008 around the invention of the iPhone and it s soon competitors. It seems to me that "the internet" took 30 years from creation to integration into society. AI programming has been around for decades but its integration into society seems to be more parallel to mid to late 1990s. At this point, its still mainly in the toy stage of development. We haven't gotten to the integrated into society section. I do not know what integration into society will look like but we found a great new tool we just haven't figure out how to use it correctly yet.
I think Google search came on line in 1998. I'd say the equivalent point for AI was at least two or three years ago. Add to that AI is being adopted for more things by more people at a faster rate so I think your estimate of a comparable timeline is off a bit.
"By 2000, it was still an unusual toy for most people.“
By 2000, roughly 50% of the U.S. population was using the www. That seems to be far more than an unusual play thing to me. Even my tech ignorant parents had their first dialup account in 1997.
Likewise, the consumer usage of AI in the U.S is roughly 50% as of 2026. Enterprise usage is something going on like 90%.
As such, I’m not really buying into your timeline or this: “At this point, its still mainly in the toy stage of development. We haven't gotten to the integrated into society section.”
And don't almost all college and even high school students use AI all the time?
the innovation is moving from the models to the harnesses, which are created by the users - individual and business. i'm a heavy user of claude.ai to manage my investments, including inputs from finance, economics and geopolitics.