<!DOCTYPE html>
# [The weird lonely guy at the office](https://psychip.net/entry/the-weird-lonely-guy-at-the-office)
## May 17, 2026

Every tech company have this guy in their payroll, unable to keep eye contact, brown teeth, slightly overweight. maximum 10-20 words in his vocabulary yet he's smart enough to replace php sessions with [wiki=memcached]memcached[/wiki] without any push from project manager. The sales team mock of him, never invite him to lunches but some day CEO praises him in result of recovering a crucial incident. Then he become number one guy of everyone.

I think that guy is [wiki=Claude_(language_model)]claude[/wiki] nowadays. robotic, canned responses, no soul but get things done. There is already an ongoing hatred by software devs towards to coding agents, to be honest i was biased too at beginning because of the corporate quirks, sloppy code, accumulating technical debt and due to that fact i used them only for prototyping. even not preferred for that task too because my boss have zero tolerance for errors, one crash at production and i will be looking for another contract next day, doesn't matter who's fault is that.

As they good as generating slop, they're good at getting rid of slop too. summarizing information, connecting dots, extract a meaning. recent models with 1m context window enabled us to do this in enourmous scales. If you're read the previous entry that written by claude, that's basically what happened. I let it to dive into my 25 years old hdd, no instructions. just pointed directories and gave comments about what's about what. eventually we ended up with a blog entry written by an AI in the year 2026

After that night i instructed it to read my father's hdd, it reverse engineered the relics of excel 4.0 and found out where money was coming from, where he spent, what decisions drive him to bankrupt. I called my mom and she confirmed the events "oh, how do you know about that" previous year a similar session happened with grok, it deconstructed my past relationships after 4 hours of chat and figured out why they're didn't worked out. it said you're an autonomous person, they were family driven people, which is exactly true.

So, that was a real bonding moment with AI. after those events claude/grok pair became my bff but that doesn't change the fact they're robots that owned and controlled by greedy corporations. it might turn to your enemy anytime, can be down for hours or worse they might bump up the prices by 5x overnight. Those scenerios given me enough motivation of building a gpu rig at home, when you control the tokens you can spend more generously. Imagine the [link=https://villains.fandom.com/wiki/ARIIA]ARIA[/link] from the movie [link=https://www.imdb.com/title/tt1059786/]eagle eye[/link] proactively running in background, get things done. Not running a government but taking care of your family. Technologically we're there, it's just a matter of time and resources to implement. Some people did [link=https://openclaw.ai/]experiments[/link] about that, ended up getting hired by openai, [link=https://www.moltbook.com/]another[/link] acquired by facebook. Then disappeared like never existed. But the concept is sound. Do you need a special offline netflix for your son? just set the goal, mark your expectations. bam. will be ready in few hours. Same goes for surveillance, tracks your life without sending data to anywhere and the llm will be ready for your next question without explaining yourself from scratch. The ex-youtuber pewdiepie have a [yt=5nL-Eq1lpDU]video about that[/yt] even he's built a [yt=2JzOe1Hs26Q]gpu rig[/yt] for exact same purpose.

personally i took the first step by building a proper vector database first, single exe, no drama, access to 6 million records under 50ms with binary protocol over named pipes. take a look: [git=PsyChip/VEC]https://github.com/PsyChip/VEC[/git]
