New chatbot AI companions for poorly socialised agentic developers

For immediate release:

ATLANTA, Georgia (1 April 2026)--Evariste Systems are pleased to announce a fully
inked partnership with the Kamailio project to provide a specially fine-tuned
AI chatbot companion for poorly socialised developers who submit AI-generated
pull requests to open-source projects.

The problem of AI-generated pull requests (PRs) has become endemic in recent
months in many telecom-related open-source projects, including Kamailio and
Asterisk. These pull requests consist of minimal developer involvement or
comprehension of the code submitted, and proffer a narrative form widely referred
to as "AI slop", consisting of signature large language model (LLM) authorial
traits such as emojis, inaccurate or overly general bullet points, emdashes,
trite clichés, often adopting a cloying, hokey tone.

Alex Balashov, chairman of Evariste, was on hand to outline the novel findings
of Evariste's ongoing research programme into this peculiar phenomenon.

Balashov first recapitulated the well-known problems with AI slop PRs for the
benefit of a more general audience:

"Contributions to open-source projects, like code contributions to any software
project, go into a living organism and must be properly metabolised.  Most
challenges of software development are not code-shaped. Given a practically
infinite ability to generate code, the bottleneck is unquestionably human
bandwidth: what are we doing and why? If we do not ask these questions,
'vibe-coded' projects degenerate from the accumulation of design entropy and
errors, hallucinations and so on.

In addition, there is of course the problem of accountability for code created
outside of one's mental control, as well as subtle incompatibilities with
project practices and established culture of problem-solving, potential
intellectual property claims by AI companies in the future, and lots of other
complications. But the most basic fallacy of AI slop pushers is plugging a
fire hydrant into the water pipes of a house and expecting a fire hose worth of
water pressure from the residential plumbing of their kitchen faucet."

The problems Balashov describes have led many open-source projects to adopt
rapidly evolving manifestos or contribution policies related to AI, with a
special focus on generated patches and commit messages in which the human
contributor appeared to have minimal involvement.

As a frontier expert on the psychological dimension of this social problem,
Balashov shared the arc of clinical thinking over the past year or so:

"We have tended to think of this as an intellectual problem. There were two
camps here: on the one hand, people who thought it was just a failure of
understanding--that if we just tell them why this is bad, they'll stop doing
it, or maybe if we craft a sufficiently in-tune policy to fit the cultural
moment or something. On the other hand, there were people who thought this was
a developmental disorder. Some said it was a kind of cretinism."

Balashov paused to note: "The cretinism angle seemed the closest fit, and I
have to admit, on the surface, it's still the most persuasive. Much as the
stigma has been removed from many mental health topics over the years, we have
come around to a more nuanced and compassionate view.

This is essentially an emotional problem, an outgrowth of incapacity to relate to
normal human beings, an inability to construct an adequate theory of mind and
get inside the head of another. There is a throughline here to more traditional
pathologies of Silicon Valley technocracy, as in the idea that all problems of
politics, governance, morality, and so on can be understood as an insufficiency
of apps or code, that what hungry people in extreme poverty most need is to be
showered with iPads, that kind of thing. Just one more app, one more platform,
and all will be solved.

At root, it's a paucity of socialisation and acculturation in the digital age.
These are deeply lonely people who are first and foremost keyed to interact
with inert and lifeless machines, not people. To them, Github usernames of
fellow human beings represent something no more conscious or alive than a
jackhammer."

Balashov added abruptly: "From a comorbidity and confounder standpoint, which
impacts study design, this pathology is coeval with the much-discussed 'AI
psychosis', or what James Howard Kunstler--in talking about American-style
suburban automobile sprawl--called 'technosis externality clusterfuck.' This
makes it challenging to address. They don't understand how ridiculous these
pull requests look from the outside, that any normal person can tell their
code was spit out by Claude, and even when it's pointed out to them, they don't
see the problem. They can't hear this critique because they don't really connect
with normal humans, just the Hacker News comments section."

And where to go from here?

"Sadly," said Balashov with a heavy sigh, "where we've landed is that this
defective mentality is so widespread that we can't really put the milk back
into the cow at this point. We have to move on to a strategy of containment,
the way we limit the damage blast radius of all deeply antisocial phenomena."

Asked about seemingly obvious parallels to the institutionalisation of mental
health patients or the American carceral state, Balashov was quick to dismiss
the validity of such a comparison.

"We're not proposing to confine LLM slop-slingers to a padded room, although
many would welcome that. They are able to carry on as free individuals in a
free society. What we need is fundamentally a way to absorb their energy, to
let them spin their wheels in a way that results in the least vandalism to
open source."

Asked by a member of the business press about the Kamailio and Asterisk
connection, Balashov said that due to the close involvement of Evariste
Systems with both projects, he has unwittingly had the opportunity to conduct
retrospective studies on sufficiently large data sets of pull requests and
forum posts with AI slop.

"What makes these projects interesting is that they rely on relatively
specialised domain knowledge, as open-source goes--these aren't REST
microservices or JavaScript widgets or something--so the contours of the
ignorance stand in sharper relief. I think this is what attracted our research
interest."

Balashov continued:

"You know the Napoleon complex, of course. The thing that keeps surfacing is
something akin to a delusion of grandeur, that one is being 'agentic'. That's
what we keep finding. These people, who by and large have lacked agency with
code throughout their lives, have allowed themselves to believe that they have
now transcended the need to 'think in code', that they are 'embracing
exponentials' with 'parallel agents' and thinking about 'business
requirements'. Incidentally, on the 'thinking in code' thing, they don't seem
to have asked themselves why they call it a large language model, but I
digress. Anyway, it dawned on us that what we really need is to somehow get
these people feeling 'agentic', but take away the sharp objects.

This leads us to the companion chatbot. Its sole social interface is as a
Github commenter. I won't go into the details of the post-training or all
our LoRAs, but we have tuned it uniquely to the clinical texture of the
'agentic' delusion."

To further questions of tuning, Balashov remarked: "Look, it does a lot of
stuff that would be hard for people with relatively normal social habits, like
journalists, to understand, but it's stuff that you can't just stuff inside a
system prompt or push into context.

For one, it's laser-focused on talking in terms of 'SWE benchmarks', it does
extensive mock code reviews, simulated grateful patch acceptances, and offers
sycophantic commentary to reassure the author that they're 'really on the right
track with all these agents', and it generates what we call 'honeypot' Github
projects. We can't disclose what these are, of course, but it'll be obvious to
most people with an adequately developed medial prefrontal cortex. Naturally,
the projects are thematically tuned to attract these people like a dim light
attracts insects.  One of our staff doctors seemed to find that they respond
best to 'AI agent harnesses', whatever that is."

Do these chatbot companions do anything special to 'gameify' the experience,
we wondered?

"They do, actually!" said Balashov. "They're really big on rewarding the user
for littering the project with .md (Markdown) files. These tortured souls are
really big on 'skills', so there is a lot of thumbs-up emojis and effusive
praise for adding them to the project. In fact, one is incentivised to make as
many Markdown 'skill' files as possible. It'll do fun stuff like randomly
assign a bug to especially prolific users, and it'll say, 'bro, if u dont add a
skill for this we r so cooked frfrfr.'

It will keep them busy. This is the overriding
need.  It's like that species in Star Trek: Voyager, back in the day, the
Hirogen, if you understand that reference.  They were hunters, and above all
else, they simply needed to hunt. They were eventually given a holodeck in
which they could holographically simulate the thrill of the hunt without
actually hunting sentient beings. It's a bit like that. There's no Star Trek
like 1990s Star Trek."

Balashov also confirmed rumours that the new AI chatbot companion requires
every user to pay for a Claude Max subscription.

"We took some real incoming for this one. It's true that setting it loose with
that many tokens is planet-warming, but so-called 'agentic prompt engineers'
are going to burn the tokens anyway. We'd rather they burn the tokens and warm
the planet in these sandboxes where they can do less harm, and we figure the
compute they use for 'vibe-coding' will take away from the global share of
compute that can be allocated toward vandalism of serious code repositories."

On the future of these specialised AI chatbot companions, Balashov was notably
coy: "It depends on where they want to 'embrace exponentials' next, to be
honest.  We're sticking with Github as a paradigm for now, but nothing's off
the table.  Gambling and erotic modes pose obvious legal challenges, but do you
want to piss into the wind explaining why vibe-coded SIP stacks are bad to
someone with an advanced case of chatbot psychosis? We've moved from a cure to
palliative harm mitigation."

--
Alex Balashov
Principal Consultant
Evariste Systems LLC
Web: https://evaristesys.com, https://www.csrpswitch.com
Tel: +1-706-510-6800

__________________________________________________________
Kamailio - Users Mailing List - Non Commercial Discussions -- [email protected]
To unsubscribe send an email to [email protected]
Important: keep the mailing list in the recipients, do not reply only to the sender!