<nettime> few notes on software and LLMs

[Novica Nakov via nettime-l <nettime-l@lists.nettime.org>]

Dear Nettime,

I think it is a good idea to post this here. These are two blog posts
I wrote recently. Is anyone writing blog posts anymore? Pasted with
the links in markdown here:

title: Software engineering enters its accounting era

If you have ever done accounting, you probably know it is not much
fun. If not, here is a brief summary. There are a bunch of rules to
follow, coded into local, national, and international standards, and
many modern ERP systems have them at least partially implemented and
automated. There is not really creativity in the work -- well not
unless you want to work in organized crime, and the profession is
highly regulated. There is a lot of gatekeeping about who can do the
job, or who can do which part of the job, and what kind of
certificates one must have, although the core of the accounting
process is largely unchanged since it was codified sometime during the
[Renaissance or
thereabouts](https://en.wikipedia.org/wiki/Luca_Pacioli).

No one has accounting as a hobby. There is no cool project in
accounting you can share with your friends, and there aren't any
interesting or thought-provoking conferences or meetups where the
topic is accounting. Was there ever an accountant speaking at TED? I
haven't checked, but if there were, he/she was probably speaking about
something else.

So what does this have to do with software engineering? To use Don
McMillan's funny diagram, we are now seeing a shift away from
problem-solving in software engineering. Instead of solving problems,
we are writing Markdown documents that tell agents to solve a problem.
And, even in these early days, we see projects popping up with default
`.md` files to do certain tasks or write in certain languages. I think
it will not be long before we see a standard `.md` file to build an R
package, or to build a FastAPI, or something else, with the complexity
of the problem being solved increasing over time.

There is a trajectory in which agent skill will become something akin
to International Financial Reporting Standards. I think we are not far
away from having the big companies producing "verified" skills (as I
went to LinkedIn to post this, I was greeted by a [post by Anton
Abyzov](https://www.linkedin.com/posts/antonabyzov_ai-developertools-buildinpublic-share-7438679148476125184-8lnk)
that talks about a [verified skill tool](https://verified-skill.com/)
:) to do certain things in their ecosystem). Think, for example, of
"AWS verified skills for infrastructure", and maybe these will start
to come with a price tag on a tier level, or even on a regional level
around the world.

Then if you are a company working in software auditing, you will
probably have a skill that will instruct your agent to audit the work
of another agent. I would not be surprised if the Big 4 are already
thinking about this.

Where does this leave software engineers? No more problem solving,
more OCD? I think it is not difficult to imagine future work being
mostly reading `.md` files and trying to catch inconsistencies that
slip through spell checkers or similar tools. Think someone is typing
`cat` instead of `car` in some `.md` file, and someone else is trying
to figure out where the error is coming from. That to me sounds much
closer to tracking the stray balance mismatch on the balance sheet
than to improving a poorly implemented function that sometimes crashes
the user's computer.

Is this good or bad? I guess it depends on being good or bad for whom
or what. I think that even if this trajectory takes place, it will
probably not affect the current generation of software engineers.

However, not thinking about the other issues being raised (such as
overall environmental concerns, employment outlook, and the stock
market bubble), I think over time we are very likely to witness the
[tragedy of the
commons](https://en.wikipedia.org/wiki/Tragedy_of_the_commons)
unraveling in the free software world.

Then maybe a lot of the software becomes like the cheap umbrellas you
buy on the street from ad-hoc sellers when it suddenly starts to rain,
and then they break after the second use and you toss them away. And
perhaps a lot of the software becomes a problem similar to the plastic
in the oceans, and [some engineers re-skill to deal with
that](https://www.theguardian.com/world/2026/mar/15/cairo-fishers-catching-plastic-bottles).

For the current generation, I think it will put pressure on community
building around software, hobby projects, and events. Think about it:
why would I want to come to a meetup if we are to discuss your latest
`.md` file? It is not meant for people anyway. Or why would I bother
with reporting a bug or offering a pull request on a project that is
largely built by a coding agent? The intrinsic motivation to help and
be involved with people is simply gone. I would even say there is no
motivation to send my agent subscription to do a pull request on your
agent-driven project. What would my lighting talk be at the next
conference: how to format your `.md` file in 10 easy steps? There is a
skill for that too. After all have you ever heard of an accounting
community?

title: Software engineering as social media

Six months ago I wrote a post arguing that software engineering enters
its accounting era. I still think this is a good argument on the point
about building software and the potential community around it.

Today I want to expand on the consequences of this new way of working,
looking at two things: decision fatigue and algorithmic development.

The term **"decision fatigue"** has popped up enough times regarding
LLMs and AI. It is about the cognitive overload of reviewing agents'
work. Should I accept this commit or not? Is this a bug worth chasing
or not? And so on.

I think it is funny the industry and/or community that burdened the
wider society with platforms that deliver decision fatigue, now faces
the same acute problem. By this I mean the way social media is
designed to work: to offer an endless stream of advice that aims to
micromanage a person's life. A user of today's social media will have
to deal with the daily information feed that the algorithm of the
platform will serve and then will have to decide on multiple things
depending on interest and context and then do it again tomorrow and
the next day.

I am making these up as I type but won't be too surprised if they are
real titles somewhere online: five reasons why eggs are best for
breakfast, three reasons to avoid eggs for breakfast, six lessons for
raising smart kids, 10 things you should never do if you want your
children to be successful, three exercises you should never skip, and
seven exercises that do more harm than good. If that sounds too
familiar, maybe it is very close to Should we chase this bug or just
commit it as is?

Then **algorithmic development** as the sibling of the above. Just as
the next thing you see on social media depends on what you clicked
before, the next suggested LLM prompt is basically a continuation of
the last output it provides. It is what the LLM is designed to offer
as a next step. That may not be the next step relevant to the thing
you are working on or even not be something you consider important,
but pressing enter is just too easy and off it goes.

Again, this comes back as a boomerang to the industry that was so
eager to design what the public consumes online in the same manner.

Some users may be better at this, of course. The same as some users
are not on social media, and they can ignore it or curate it better.
But for many, if not the majority, it will be, if not already, a crap
experience (even without mentioning the ads, which are to become a
feature of LLMs.)

And some users may enjoy building the random thing. GitHub will become
the Myspace of the LLM world with various scattered projects that
people used once or tried to make it work with the extra $20 spent on
an LLM subscription instead of a few beers or whatever. Will many of
those be things that you would like to return to or use in two years'
time? I doubt it. But that's what social media is about, right? And
endless stream of transient experiences.


-- nnvc
-- 
# distributed via <nettime>: no commercial use without permission
# <nettime> is a moderated mailing list for net criticism,
# collaborative text filtering and cultural politics of the nets
# more info: https://www.nettime.org
# contact: nettime-l-owner@lists.nettime.org