> You bought a laptop or desktop with an operating system, and it did what it said on the tin: it ran programs and stored files.
I feel like people may be viewing the past with rose colored glasses. Computing in the 90s meant hitting ctrl-s every 5 seconds because you never knew when the application you were using was going to crash. Most things didn't "just work", but required extensive tweaking to configure your ram, sound card... to work at all.
This is not just the past. I still have headaches configuring my video card to work with the right CUDA drivers, etc.
The tower of abstractions we're building has reached a height that actually makes everything more fragile, even if the individual pieces are more robust.
alot of software engineering, especially in complex systems, is still just tweaking retries, alarms, edge cases etc. it might take 3 days to even figure out what went wrong
Still though -- once you got a workflow, no matter how terrible, it strongly tended to continue to work that way, and it was still much easier to diagnose, fix, and just generally not have unexpected behavior.
This is the issue; agents introduce more unexpected behavior, at least for now.
My gut is that always on "agents who can do things unexpectedly" are a dead-end, but what AI can do is get you to a nice AND predictable "workflow" easier.
e.g. for now I don't like AI for dealing with my info, but I love AI helping me make more and better bash scripts, that deal with my info.
> Computing in the 90s meant hitting ctrl-s every 5 seconds because you never knew when the application you were using was going to crash.
THIS.
I lost so much work in the 90s and 00s. I was a kid, so I had patience and it didn't cost me any money. I can't imagine people losing actual work presentations or projects.
Every piece of software was like this. It was either the app crashing or Windows crashing. I lost Flash projects, websites, PHP code.
Sometimes software would write a blank buffer to file too, so you needed copies.
Version control was one of my favorite discoveries. I clung to SVN for the few years after I found it.
My final major loss was when Open Office on Ubuntu deleted my 30 page undergrad biochem thesis I'd spent a month on. I've never used it since.
Quality issues are a different vertical within the space of software/user misalignment. The sort of issue the author talks about is more like the malware of the 90-00s era: the software deliberately does something to screw the user.
been building on claude code for a while. the post's framing is right.
mcp gives you open standards on the tool layer but the harness
(claude code, cursor) is still proprietary. your product is one
anthropic decision away from breaking.
the user agent role the post calls for needs open harnesses, not just
open standards. otherwise we end up rebuilding mobile under a new name.
The thing I don’t like about “agents” is that I consider my computer a tool that I use and control. I don’t want it doing things for me: I want to do things through it. I want to be in the driver’s seat. “Notifications” and “Assistants” and now “Agents” break this philosophy. Now there are these things doing “stuff” on my computer for me and I’m just a passenger along for the ride. A computer should be that “bicycle for the mind” as Jobs put it, not some autonomous information-chauffeur, spooning output into my mouth.
i think whats missing is the raison detre of the Agents isnt a new usecase, its a context prune for the same limitations LLMs provide. LLM as Agent is a subset, where the goal of the agent is set by the parent and is suppose to return a pruned context.
if you dont recognize the technical limitations that produced agents youre wearing rose tinted glasses. LLMs arent approaching singularity. theyre topping out in power and agents are an attempt to exentend useful context.
The sigmoid approacheth and anyone of merit should be figuring out how the harness spits out agents, intelligently prunes context then returns the best operational bits, alongside building the garden of tools.
Its like agents are the muscles, the bones are the harness and the brain is the root parent.
I feel like people may be viewing the past with rose colored glasses. Computing in the 90s meant hitting ctrl-s every 5 seconds because you never knew when the application you were using was going to crash. Most things didn't "just work", but required extensive tweaking to configure your ram, sound card... to work at all.
That was in the Windows world. Maybe in the Mac world too?
No so much in the *nix world.
Windows seems to have improved its (crash) reliability since then though, which I suppose is nice. :)
I used computers back then and many things just worked fine. I found Windows XP way more predictable and stable than any of its successors.
The tower of abstractions we're building has reached a height that actually makes everything more fragile, even if the individual pieces are more robust.
Have people outgrown this unnecessary habit? Haha
This is the issue; agents introduce more unexpected behavior, at least for now.
My gut is that always on "agents who can do things unexpectedly" are a dead-end, but what AI can do is get you to a nice AND predictable "workflow" easier.
e.g. for now I don't like AI for dealing with my info, but I love AI helping me make more and better bash scripts, that deal with my info.
THIS.
I lost so much work in the 90s and 00s. I was a kid, so I had patience and it didn't cost me any money. I can't imagine people losing actual work presentations or projects.
Every piece of software was like this. It was either the app crashing or Windows crashing. I lost Flash projects, websites, PHP code.
Sometimes software would write a blank buffer to file too, so you needed copies.
Version control was one of my favorite discoveries. I clung to SVN for the few years after I found it.
My final major loss was when Open Office on Ubuntu deleted my 30 page undergrad biochem thesis I'd spent a month on. I've never used it since.
There is no legitimate intermediate position - The skew will go one way or the other.
mcp gives you open standards on the tool layer but the harness (claude code, cursor) is still proprietary. your product is one anthropic decision away from breaking.
the user agent role the post calls for needs open harnesses, not just open standards. otherwise we end up rebuilding mobile under a new name.
[1] https://github.com/mistralai/mistral-vibe
[2] https://goose-docs.ai/
if you dont recognize the technical limitations that produced agents youre wearing rose tinted glasses. LLMs arent approaching singularity. theyre topping out in power and agents are an attempt to exentend useful context.
The sigmoid approacheth and anyone of merit should be figuring out how the harness spits out agents, intelligently prunes context then returns the best operational bits, alongside building the garden of tools.
Its like agents are the muscles, the bones are the harness and the brain is the root parent.