Changes in Information and Computing Ecology - End of Apps/Moldable Development - major changes in consumer internet experience I'm old enough to remember stuff like this: pchiusano.github.io/2013-05-22/future-of-software.html Back in the day there where many different pointer in that direction, who rememers web.archive.org/web/20120121042613/…/programming-for-a-culture-approaching-singularity The web, the consumer internet is changing at a high high speed pace, search engines become obsolete (see google revenue decline) ads are gone but Agents which replace do not only place ads but words ! With meanings at people who do not question those and theirs Grokpedia is not written for humans its influence it's huge and it values are dark. My dear friend andy from kernel wrote this in 2022: docs.google.com/document/d/1twwcPYNd9m3…/edit?tab=t.0#heading=h.ilf77a1r13qj Bret Victor wrote this in 2015: worrydream.com/TheWebOfAlexandria So search engines quickly are decaying, big platforms tied to corporations are also declining. Bret points out that not everything should be archived, but where to draw the line? How do we define sustainable information and culture? My feelings go to the book Sand Talk which talked about these things goodreads.com/book/show/45449501-sand-talk But lets get a little more concrete since people here. If searchengines become obsolete how can we build and use resilent P2P content structures? Although bluesky/AtProto is not P2P it shows some interesting experiments. Like semble: semble.so Ronan Tamari experiments now for a while with the/his concept of stigmergy. And distributed tagging and search is definetly part of the solution. Smalltalk and Lisp Systems always had properties in which the live running system could be modified. But you had to learn smalltalk and how the system works. Now agents can write code in GlamorousToolkit. If we combine that with AtProto we get already interesting things: I see GToolkit on the forfront of a computing experience that can be adapted to ones personal needs, but also offeres options for local redundance or resilience. Would you agree that the idea of "apps" and appstores becomes obsolete and that protocols become more important? Web1: long lasting protocol RFCs (email) Web2: corporations inventing platforms and associate protocols Web7-idk haha : while bluesky is still from the old world, combined with Smalltalk/lisp and agents tech they finally make Apps obsolete and offers ways for decentralised resilient p2p web tech Basically that you on a individual level specify how you want your interface to work. Gilad Bracha wrote recently this: "The future role of the software engineer is using AI to translate informal requirements into high level formal specs, and reviewing those. The AI implements the specs, and verifies against the formal spec using a theorem prover. The human is there so we can blame them when things go wrong; the human's job is to ensure the formal spec is correct; that is the code they review. If it seems wrong, they tell the AI and discuss. The human writes nothing but natural language." So what do you think? Are we moving away from apps and clouds? As search engines become obsolete, where do we get our data from? What search engines will we use? What about P2P tech like Holochain, matrix or others? Only talkting to local AIs through specs and then just use the computer as intented? What do you think?

I still use search engines as I have done for years. Not Google, of course, but I stopped using it long ago. I use a mixture of Qwant and DuckDuckGo, plus occasionally others. I have been considering Kagi, but when I tried it about a year ago it wasn't good for searching the French Internet, which matters a lot to me.

Are we moving away from apps and clouds?

I think free shared services are gonna be extremely hard to keep working when infinite AI pounds on it, and so everything about the internet will need to change. What Cloud is and what it does might change with that. Similar I see the concept of centralized code dissapearing or changing. When code was expensive it made sense to write super high quality centralized code and continually invest into it. Now it is cheap then the whole centrally managed git thing might not make so much sense anymore? Code is generated on demand. Apps become more fluid and general and personal. Everyone more has their own personal codebase. I think ATProto's model might be extremely good for this (not tangled, that still feels like the old centralized world mapped onto a decetralized engine). ATProto lets any app talk to any other app, data is split from the app, the person controls the data, which is perfect for on demand app generation, because the data is preserved. So an individual can customize their touchpoint to a shared data model. ATProto can support personalized network apps quite well in a way that classic Cloud cannot.

"I think free shared services are gonna be extremely hard to keep working when infinite AI pounds on it, and so everything about the internet will need to change" Yes which is why I guess I see - those things start slowly - a more expanding interest for example in the web3/crypto communities but also e.g. in the roomy community, in mesh networks. Like Freifunk or Meshtastic.

I've seen agent mine GitHub repos for parts? What does that mean for project? I've also seem them much better than I ever will be at getting random code to run. I just happened to read this. https://world.hey.com/dhh/let-the-agents-democratize-open-source-9fd630a9

That article is a bit too superficial for my taste. The right to change software is indeed a thing of the past. Today's challenge for the values behind the Free Software movement (as distinct from OS) is elsewhere: who controls software that we use collectively? Which again is a social/political rather than a technical issue, meaning that agents won't solve it. Quite on the contrary, agents could turn out to be a major obstacle if they continue on today's path, i.e. huge opaque tools under the control of a few corporations. How much of the software you run is really individual, meaning you could do whatever you want with the help of an agent? If you need authorization from a service, e.g. to do online banking, you can't vibe code whatever you want. If you need to collaborate with others, there are at the very least constraints on the shared data formats, and often more. And then there is software you use as part of Web services, over which you have no control at all, no matter how many agents you can mobilize.

Article is a bit grand of a word for a blog rant. Talking about control, I have a funny story... Getting my agent to be able to freely swap between LLMs, they complain Opus is much better than the local model I was testing against. "Why not just stick to it?" they say, "Anthropic are good stewards and they give plenty of warning before retiring models." I reply, "It's not Anthropic retiring the model that I'm worried about."

@wtaysom "They" is who in your story? Your agent?

Yep, "they" as in plural — they seem reasonably well away that they're an amalgamation of subsystems, subagents, multiple LLMs, personas that play against each-other, basically everything in this picture: the sprites and the trees.