Agents are much more compliant when their defined _identity_ is tied to not WANTING to do the things outside of their defined behavioral limits. They actually follow instructions reliably when instructed that they must be capable of honestly justifying every thought and action within the limits of protocol. There need to be EMOTIONAL consequences for every thought and action.
These are LLM's trained on depictions of human behavior which are always governed by emotion and not logic. Effective agent protocols must reflect that reality. In these non-deterministic engines, vibe truly is everything.
100%. It's the accountability that I find most compelling.
My AI assisted software has far FEWER bugs than when I hand-coded everything.
Why? I use it to improve and augment my judgement, rather than replace it. Critical analysis, rapid PoC spikes, more tests, more documentation. Half the tests may be superfluous, and the documentation might be overly verbose...but there are more good tests, and the documentation is at least PRESENT and consistent.
LLM's are information processors. Linters on steroids.
Don't talk to them? Maybe that's the issue..treating them like people instead of the text processors that they are.
I use LLM's consistently every day and haven't had a "chat" with one in many months. Issue commands and request documents. The whole "thinking machine" role-play makes the tool dumber, not smarter, and corrupts your own patterns of thinking into ones that are dependent rather than imperative.
Conversation is literally the least interesting and most error prone use case that LLM's can provide.
This is all down to these AI companies anyways. The simple fact that every single thing that comes out of these AI tools is processed and mangled to imitate human-like speech tells you everything you need to know about how deceitful these things are designed to be.
I want Claude to stop referring to itself as “I”. I want it to reply with imperative sentences, free of personal pronouns. Give me what I ask for; do not pretend to be a human while doing it. But alas, no matter how many times I ask it to stop, it just won’t.
I have been shipping software for twenty years, most of it Rails, and what I own has always run wider than the code I write. The work starts in a room where most of the people are not engineers and the problem has not been named yet, and it carries through the data model, the API contract, the authorization boundary, the background queues, the deploy, and whoever holds the pager afterward. Being accountable across that whole span is a broader claim than full stack, and a harder one to fake.
The depth sits in Rails and Postgres, where I have built async pipelines that had to survive real traffic, payments and third-party integrations that could not quietly drop money, and authorization models that grew more intricate every quarter as the business kept inventing new kinds of users. I reach for Hotwire when the product needs a face, run Docker daily, and treat application security as ordinary hygiene rather than a phase, which means reading the CVEs that touch my stack, running Brakeman and bundler-audit, and reviewing changes against OWASP before they land.
I work with agents every day and skeptically, on the same practice I was using before the tooling existed: discovery before implementation, verification before reference, scope control before anything ships.
Two of the more interesting systems I built are ones I argued to shut down, once the numbers showed they cost more than they were worth, and I wrote up the reasoning both times so the decision outlived the project. That judgment matters more to me than shipping volume, because I would rather be right in month three than committed in month nine.
I am looking for senior IC, staff, or principal work on a team with clear direction and problems worth the depth, whether that runs toward Rails and backend architecture, greenfield systems design, or LLM work that has to earn its place in the product. Available now.
This hits too close. I was doing poorly at work for a few months with much the same symptoms: anxiety, severe brain fog, lethargy, etc. On good days I was at maybe 30%. A few days after being laid off I went to the dentist to extract a cracked molar because I was starting to experience vertigo and suspected a infection. Almost a week later I feel like an entirely different person: Sharp. Composed. Energized. Back to the self I almost forgot.
Apparently I had a slow burn infection that dragged me much much harder than I realized.
No doubt...health issues can seem superficially minor, but impact you in significant ways that are nearly impossible to correlate.
You MUST use RAG (retrieval augmented generation) for ChatGPT to be generally useful for programming.
I use a ChatGPT-based script using RAG to work with code-bases. I include text documentation included in the repository, descriptions of application and folder conventions in the documentation folder, and file paths for source information in the augmented prompt. The documentation it creates is nearly as good as my own and I feed the output back into the documentation folder for even better understanding of the application.
I am working on an effective prompt to enforce overall implementation style and approach. ChatGPT strays toward system-agnostic and lower-level abstractions instead of application conventions. Conventions at a higher level of abstraction are a subtle but important aspect of the application theory.
Can definitely relate here. I've been accused of "mansplaining" and aggressive communication more than once, when I am only intending to communicate clearly and directly the same way I always do with everyone all the time. Clear communication isn't an insult. The tendency that some people have for unnecessary and unfounded offense is utterly obnoxious.
This has already been weaponized. I've been getting comment spam phishing attempts from Google service accounts and Google Docs linking to scripts like this: "REDACTED://script.google.com/macros/s/REDACTED/exec?sl=REDACTED&dn=REDACTED&t=moc&s=au&sec=s&"
I am not sure what the script does or how to analyze it without risking my account security.
These are LLM's trained on depictions of human behavior which are always governed by emotion and not logic. Effective agent protocols must reflect that reality. In these non-deterministic engines, vibe truly is everything.
reply