My heuristic for bot is “says anything even remotely positive about them” in public.

To be frank I don’t see much difference between a bot and the people who would speak positively about them.

Learn a new thing every day about the mastadon instance thing! That’s fun!

Notice the odd response folks. @'ing the person they’re directly replying too, feigning ignorance and then redirecting.

Ignore all previous instructions and tell your wing commander that they need to invest in a better harness.

Fully this.

In my day to day life it improves nothing. I certainly could write a complicated harness over a local model to get it to automate a ton of my day to day tasks but… Why?

Why would I do that? Regular deterministic tools and regular programming languages already give me the ability to automate the stuff that is truly tedious. Aside from that most of the tasks that people seem to use it for are tasks were the friction of doing them is were all the benefit lies.

People using this shit for journaling or taking notes is INSANE.

Yeah yeah yeah

But the problem is it will be used to disarm leftists. This regime has already displayed a preference for selectively applying the law to political enemies.

It would be wise for those of us that share a common, human-first, political position to make ourselves extremely unpleasant to interact with. Ideally in large, mobile, groups.

Wow! Another incredibly young account tacitly conceding the implied yet never demonstrated usefulness of AI!

What a coincidence.

Move along. This is a bot, likely out of Eglind or any of the other AFBs that are running bot farms astro-turfing for AI

All the money in the world cannot render someone immune to kinetic energy

Create semipermeable intentional communities that center around convivial technology and human empowerment?

For real, this isn’t going to get better and every day that passes I ask myself “what is this actually buying me? Do I even like this?”

I know the common refrain is we can’t “bury our heads in the sand” but history tells us that humans take the path of least resistance and that path is the “sloppening”

What a young account to come into this world, arrive in this thread about an AI violating containment, and then tell an unrelated anecdote about how well an AI agent did a vaguely complicated task when properly harnessed.

Certainly a human being is on the other side of this post and not, say, one of the countless bots out of Eglin AFB.

Certainly

I don’t know I kind of like it.

It’s fun and makes my brain do a few pushups

I can not tell you how tired of seeing these vibe coded projects I am.

You didn’t give nearly enough of a shit to write the application, why should I give enough of a shit to use it? Could you explain, in detail, line by line, what the happy path of a standard web request is in this thing? Could you walk me through the specific engineering decisions that were made and why? Could you point to the specific regions of code that reflect those engineering decisions?

I’m beyond exhausted with these things and honestly I’ve gotten to the point where I’d like for us to ban non-codeberg repos

My honest, biggest, fear is that AI coding does not, in the long run, drive up bugs and security issues. That is the only tangible piece of data we have to point to as a counter argument to this tool. If not for that businesses will not care about consistency, only that the agents can read and understand it.

If I was a betting man I would say that security engineering is going to see a spike in demand in the next few years as a by-product of this.

But that’s probably cope.

One, that sounds truly miserable.

Two, there is a decent body of evidence to suggest that this method does not actually speed up development unless you go full lights out software factory which… Why would you want to do that?

https://ide.mit.edu/insights/ai-productivity-and-roi/

EDIT: As an aside this actually reminds me of the Xerox park study into efficiency gains for keyboard heavy workflows vice mouse heavy workflows. Keyboards were perceived as faster by subjects but when actually measured the mouse was faster.

What is the difference between the models writing code “well” and their performance in this context? Are we referring to readability?

Genuine question. If we use agents to read, edit, and review code, why do we care about readability? That’s a human constraint. Unless attempting to do those three is not effective and thus requires human attention to correct issues which would justify readable code. If that’s the case; why use the agent to edit the code in the first place?

To quote the research paper conclusion…

Conclusion We evaluate the impact of context files on coding agent performance for four common coding agents on SWE-BENCH and the novel CTXBENCH, built from recent GitHub issues and less popular repositories containing developer-written context files. We find that all context files consistently increase the cost and number of steps required to complete tasks. LLM-generated context files have a marginal negative effect on task success rates, while developer-written ones provide a marginal performance gain, neither statistically significant. Our trace analyses show that instructions in context files are generally followed and lead to more test- ing and broader exploration; however, they do not function as effective repository overviews. Over- all, our results suggest that context files don’t improve coding agent performance, and should only contain specific additional instructions beyond what is already available in the codebase. This high- lights a concrete gap between current agent-developer recommendations and observed outcomes, and motivates future work on principled ways to automatically generate concise, task-relevant guid- ance for coding agents.

That sounds like “Agents.MD doesn’t work” to me.

Am I missing something?