I experienced a similar interaction recently, where this principle was hard to apply, when I was emailing with a CTO / hiring manager who had some "deeper" screening questions. It was essentially:

1. HM: AI generated email with "tailored" questions

2. Me: AI assisted response with answers (I confess)

3. HM: AI generated email with a "thoughtful" response + invite

4. Me: AI generated "thank you & looking forward" response ...

Looking back at the thread, I have to laugh and cry at the same time. It's so obvious and sad.

I lost a camera recently and set up a Claude Cowork daily scheduled task to check if it finds my camera on a few marketplaces - Claude also goes through all the images and checks if it can see the serial number, which is engraved in the body. No luck so far, but we keep trying! It was incredibly easy to set up and uses my Chrome browser so it's logged into most platforms already.

would be a crazy story, if it works out. sorry for your loss

Reminds me of a few parallels, mainly the attention economy [0] and The Social Dilemma documentary [1]

[0]: https://en.wikipedia.org/wiki/Attention_economy

[1]: https://en.wikipedia.org/wiki/The_Social_Dilemma

And what does the iPhone have to do with that? What are the hypotheses? Can't tell from the abstract at least

Definitely less mysterious to what we’re made of (-:

Lock-in / switching costs are increasingly concerning me. I am using Claude for a good year now and have been accumulating so much "knowledge" in there by now. If Claude became less favorable in terms of price/performance in the future, that would worry me. I've started to think about a distributed solution, where my storage is detached from the inference, but currently Claude is still the way to go for me. Wondering if anyone has similar concerns?

Isn't all the "knowledge" just text files? I've transitioned between services easily by simply copying the text files.

You can even just instruct the LLM to create a context file for you! They are surprisingly good at that as well.

Studies show that LLM-generated context files have a negative impact on LLM performance: https://arxiv.org/abs/2602.11988

This.^ I realized this first when moving a design spec from Claude chat to Claude Code and panicked. I literally had to build something like Notion but for agents to act as a portable memory between all cloud and local models and agents. But honestly it paid off!

If you are interested you can try it out at markbase.cloud (disclaimer and all that). I am not charging for it.

We run a "context" repository that enables us to transition pretty seamlessly from model to model (usually codex to claude and back). It has skills / plugins / connectors / tooling in relatively malleable MD files. That's what I see as the future. Rather than exporting IDE settings we'll just carry our markdown to the next best tool.

It's hedging a bet at this point, but that's why people say there's no moat. If the tools are properly used + maintained, there should be no reason we can't use a new provider even next week (maybe with a little tweaking).

that's an interesting approach and something i also considered (using git to avoid conflicts). one thing i needed was a "database" (basically a folder of markdowns) with a fixed schema so i can let the agents record their decisions in (for example when the code conflicts with product design spec). this combined with search has been a real lifesaver.

this is how it works: https://help.markbase.cloud/humans/collections/overview

Believe it or not, after writing this comment I was doing some more reading on the task. I'm planning to reorganize our context repo after finding this paper (it argues that AI generated context files can stunt the performance of models):

https://arxiv.org/abs/2602.11988

For what it's worth, if you were considering building context out.

Very interesting. Anecdotally I’ve found the opposite to be the case. But I’m very interested in understanding more. Thanks for sharing

What knowledge?

Unless you work in some obscure domain, chances are that any general "knowledge" Claude has "learned" is already public data somewhere.

If you don't believe me, launch Codex and immediately start working on the same project (s). You might discover that all the knowledge accumulated means almost nothing.

Claude Code definitely remembers things about you. For just one of the more obvious examples: I was recently asking it to make some suggestions on software alternatives, and part of the answer included (paraphrased) "While a hosted service may be attractive due to your small ops team size, your experience with hosting Linux container-based services puts this squarely in the realm of an option for you." My prompt mentioned nothing about this.

This isn't something that is public knowledge, in the sense that you mean it.

Just earlier today it asked me if I wanted to create a jira ticket for something I asked it about doing. My prompt mentioned nothing about jira.

If you use Claude Code, you might want to take a look at the "auto memories" files that it creates. See "/memory" for some more information.

My favorite solution to this is to use the Cline coding agent, which is open and allows you to easily switch between different providers and models.

Knowledge in there?

Where is the knowledge stored?

All of my knowledge typically gets stored in plans outside of the agent?

And each agent window gets archived regularly, anyways.

Not worried at all. Switching is trivial. Rebuilding context isn't very difficult and harnesses are a dime-a-dozen.

I'm curious as to what they could have done at what point in time to avoid this? Or was it "unavoidable"?

Apple managed to stay ahead of the competition for years now, but I guess the iPhone and Macs are just more complex products and revolve around an ecosystem, so not really comparable I guess. Still, I'm fascinated by the GoPro "case". I had a Hero 3 back in the days and loved it!

I think comparing companies to Apple is kinda silly as they're unique. They're in a position of such market dominance that they can afford institutional arrogance.

However, it's very illustrative in this situation because I think GoPro made the mistake of thinking they were Apple, and at their outset they were compared to whatever other dross was on the market. Unfortunately it went to their head (almost certainly they were run by MBAs instead of product people or engineers, or anyone who would actually use an action camera seriously) and they completely lost sight of what their product even was.

I think it's completely avoidable, they could have just kept trying to make good products instead of whatever subscription service bullshit they attempted. They saw they had market dominance and so they tried to rape their customers as best they could because they didn't see the customers as having a choice. Don't do that. They cashed in on their market position for short term gains and it destroyed their future.

Curious what defines a retro camera for you? You mean a digital camera that imitates some analog feel?

> DNA and evolution, even with billions of years to think about it, is really a bit of a beginner when it comes to protein design.

I like how you say evolution is able to think when in reality it's just a mysterious function of variation, selection, and time.

I find it completely daunting to speak of evolutions processes without some anthropomorphism sneaking in, despite being a hardcore atheist.

It's all so complex, and our verbs that more literally describe the billions of nanosecond operations going on in the cells feel inadequate. "When a protein molecule in an appropriate folded shape and orientation happens to be bounced by kinetic energy into the attractive region of a corresponding protease..." versus "The protease grabs the protein and cuts it into..."

Back in the days the Chinese used to copy from the German automotive industry. Now it's about time the Germans start to learn from Chinese engineering.