I totally missed you can embed AI features directly into claude artifacts. I ran the first GPT store hackathon in NYC and I feel like people were very underwhelmed. What could you do in a custom GPT you couldn't with a shared chat? The integration felt very annoying. Claude I feel nails how easy it is to build + how easy it is to share
I'm guessing the OP is looking at this through the lens of more mom and pop users on the chat interface rather than the greater product line of these companies with all their dev tooling.
I’m somewhat indifferent to artifacts. While they can produce useful code occasionally when the codebase is relatively small, they become excessively cumbersome once it reaches around 500 lines. This makes it difficult to even add or remove code. If you want to change a few colors in a webpage, they have to rewrite the whole artifact. It frustrates me that you can’t edit a single line in an artifact; you have to rely on the LLM to do it.
> OpenAI appears to have basically abandoned Custom GPTs since their Spring ‘24 update, and I’m a bit stumped as to why.
It simplifies the product, reducing the number of hurdles the user has to jump through. "Hmm, which gpt should I use for this task?" That should be OpenAI's problem, not mine!
Hilariously, today, I unsubscribed from claude due to the increased api timeouts and expensive usage costs when I can use other models for way cheaper that perform equally as good.
I'm skeptical of Anthropic's product approach with Artifacts. They made a real winner with claude code and their subscription decision (basically giving up on the profitable api metered billing to win over the professional user base) for this product. Otherwise they are making now the same mistakes Openai did with the GPT Marketplace. As a customer i don't want millions of half-baked ai mini apps to choose from. This is what i'm getting with Artifacts and those gpt stores - decision fatigue and disappointment. I want the best app to solve my problem - this almost certainly won't be these hobbyist custom instruction wrappers but well designed apps by professionals (who know the limitations of these platform won't allow them to build the best-in class solution). Just like Claude Code isn't an artifact but a complete standalone app that doesn't fit into the limitations of such marketplaces.
Artifacts can reduce token usage by a factor of 2 or more.
I had an [agent evolution framework](danieltan.weblog.lol/2025/06/agent-lineage-evolution-a-novel-framework-for-managing-llm-agent-degradation) before that dumped the output analysis into chat. It often timed out before the 10th conversation. After dumping the analysis into an artifacts, and have the LLM only edit it as required, I can go to 15 or more rounds without hitting the context limit. While they seem to re-output the entire artifact each time, they don't actually consume the tokens for the entire artifact.
This also greatly reduces the tendency of HALO-style rampancy, or AI psychosis which is also what the recent paper on context-rot/poisoning (https://research.trychroma.com/context-rot) is about.
Custom GPTs are a meme product anyway. It’s just someone else writing a prompt that you can’t see. There’s no fundamental difference between the default model and the custom GPT, so there’s no real point to them.
If they were fine-tuned models then it would be interesting.
"OpenAI appears to have basically abandoned Custom GPTs since their Spring ‘24 update, and I’m a bit stumped as to why"
My guess is it was always meant as marketing scheme, having others promote OpenAI for free, while at the same time they would be able to see which topics got traction and which not, providing potentially valueable business development clues.
I think it turned out both of these turned out not to be as valueable as they hoped.
16 comments
[ 3.9 ms ] story [ 53.7 ms ] threadI'm guessing the OP is looking at this through the lens of more mom and pop users on the chat interface rather than the greater product line of these companies with all their dev tooling.
I turned off artifacts months ago because it would:
- frequently update code incorrectly / bad edit diff
- act like it updated / created an artifact when it just did nothing
- slowly / painfully delete every single line one by one before rewriting
- use artifacts for things that shouldn't have had any code written at all
Just wasn't worth the value it provided. This was before claude code.
It simplifies the product, reducing the number of hurdles the user has to jump through. "Hmm, which gpt should I use for this task?" That should be OpenAI's problem, not mine!
I had an [agent evolution framework](danieltan.weblog.lol/2025/06/agent-lineage-evolution-a-novel-framework-for-managing-llm-agent-degradation) before that dumped the output analysis into chat. It often timed out before the 10th conversation. After dumping the analysis into an artifacts, and have the LLM only edit it as required, I can go to 15 or more rounds without hitting the context limit. While they seem to re-output the entire artifact each time, they don't actually consume the tokens for the entire artifact.
This also greatly reduces the tendency of HALO-style rampancy, or AI psychosis which is also what the recent paper on context-rot/poisoning (https://research.trychroma.com/context-rot) is about.
If they were fine-tuned models then it would be interesting.
My guess is it was always meant as marketing scheme, having others promote OpenAI for free, while at the same time they would be able to see which topics got traction and which not, providing potentially valueable business development clues.
I think it turned out both of these turned out not to be as valueable as they hoped.
Is this supposed to be a compliment?