Why is AI leading to so little innovation?

By Artemis

Made with by a Human

Certificate 7gU3D4HnCdzXcvkh
Analysis

Each score must exceed its threshold to pass. A minimum number of passing scores is required for an overall pass.

Internal Paste Percent
3%
Pass

Internal Paste Percent

Measures the proportion of your final document that was pasted from within the same document. Some internal copy-paste is expected — it reflects a human writer reusing and reorganizing their own earlier text.

Pass

A sufficient amount of the document originated from internal copy-paste.

External Paste Percent
2%
Pass

External Paste Percent

Measures the proportion of your final document that was pasted in from an external source. Too much external content can indicate the text was not primarily written here.

Pass

The proportion of externally pasted content was within acceptable limits.

Delete Percent
21%
Pass

Delete Percent

Measures what fraction of your editing activity involved deleting text. Human writers naturally delete and revise as they write. A very low delete percentage can suggest the text was not written here organically.

Pass

Enough editing activity involved deletions, consistent with human revision.

Active Typing Speed
79 chr/min
Pass

Active Typing Speed

Measures your average typing speed during active work sessions. Typing speeds that are unusually fast may indicate automated or programmatic text input rather than a human typing.

Pass

The typing speed was within the range expected for a human writer.

This article was written in  56 minutes over the course of 53 days. 89% of the text was typed and 11% was pasted from the clipboard.

Upd: Jun 27, 2026

A CEO at an AI company recently said something that resonated with a lot of people. Talking about AI coding tools: "Your org rarely has good ideas. Ideas being expensive to implement was actually helping."

I've been wondering recently why LLM coding tools seem to have had so little secondary impact. I mean, the coding tools themselves seem popular and, even if many people don't like them, there are thousands of people claiming that they're shipping at 2x, 5x, or 10x increased rates.

So where is all the stuff they're making?

The pro-AI folks have been 10x as productive for about a year or 2. That's 20 years worth of software development.

...so... where is it?

I have no doubt that the code was actually generated. I'm even willing to take on faith that that code runs and does something. But the real bar is for that software to do something useful for people. And IMHO it should do something non self-referential.

Yes there are now mountains of apps that will help you use AI to generate X or Y. There are vibe coded apps to help you improve your vibe coding. There are agentic workflow platforms, marketing content generators and so on.

I see a lot of existing apps adding AI assistants. You can use AI in google docs, Figma, outlook, etc.

But these are all tools that help you use AI. At some stage all those tools that help you use AI should help you do something other than just using more AI.

And this is the thing I'm failing to see. I have not seen a noticeable (let alone 10x) increase in the quality of the software I use. I have not seen an increase in the number of useful apps on my phone or in the number of fun games to play.

Buying something online has not become easier, not on Amazon nor on the website of my local super market, for example, by helping me search or helping me compare products. Yes, I'm aware Shopify created an AI shopping assistant that does that. The issue is that this tool has not been integrated everywhere. If the cost of building the tool is now nearly zero and the time to implement is nearly zero, good ideas should permeate the market nearly instantly.

I get that flying cars invented by AI are a few years away but software should move at the speed of light electricity.

The answer is probably that building, shipping, marketing, and selling products is still hard. I've never met an engineer, even before all the AI hype, that told me that writing code was the hardest part of their job. Back in 2021, some people suggested less than 50% of developer time is spent on coding. Others put the number at one hour per day.

Sure, coding was a point of friction, but not the only point of friction. If you completely eliminate all time spent on coding, all you're really doing is improving production to the maximum rate of the next bottleneck. And there are many.

The reason we're not getting a better Google Maps competitor, or a better Amazon market place is because these apps benefit from massive amounts of private data that competitors don't have.

The reason we're not getting better public transportation or cheaper healthcare is because these things rely heavily on physical infrastructure and in-person contact that LLMs don't help with.

The fact that building apps is a little easier now and more accessible to more people is ultimately a good thing. But making the building easier also adds noise. Now that (supposedly) everyone can make software, it's all the harder for the good software to get any attention.

Popular Posts