Building Pi with Pi
A new open source LLM “agent harness” called Pi is prompting debate over how developers should interact with AI coding tools and whether it’s worth building custom harnesses versus relying on mainstream, better-instrumented platforms. Commenters highlight maintainability problems caused by LLM-generated “slop” in bug reports, arguing for stricter issue formats and clearer invariants as agents take on more development work. The project’s use of the term “clanker” for AI systems also draws scrutiny, raising broader questions about anthropomorphizing software, inventing slurs for machines, and how such language might shape both human attitudes and future model behavior.
Font in the screenshots
- Multiple commenters try to identify the code font; consensus lands on Berkeley Mono, with minor glyph differences noted.
- Author of the image confirms using Berkeley Mono, plus other monospace fonts depending on mood.
“Clanker” vs “Agent” and Agency
- Large part of the thread debates the term “clanker” for AI systems.
- Some see it as a useful, less corporate-sounding alternative to “agent,” emphasizing that real agency belongs to humans.
- Others strongly dislike it, describing it as an ugly, deliberately derogatory slur that evokes negativity and even prevents them from finishing the article.
- Several argue that using slur-like language for machines is unnecessary and risks normalizing slurs generally, even if machines have no feelings.
- Others mock the idea of “offending software” as peak over-sensitivity and insist it’s fine to insult inanimate objects (cars, tools, etc.).
- One subthread notes Wikipedia now describes “clanker” as explicitly derogatory for robots/AI, which some find worrying for future training data.
- There is broader disagreement over what “agency” means:
- One side uses a stricter, choice-based definition reserved for humans.
- Another points out that everyday definitions do allow talking about machines having “agency” in a limited, operational sense.
Anthropomorphizing AI
- Multiple commenters warn against anthropomorphizing LLMs, seeing “clanker” debates and “offense” language as a symptom of this.
- Others mention sci-fi and roleplay contexts where “clanker” has already been treated as a slur and even banned, reinforcing the slur framing.
Agents, slop, and tooling
- Some discuss “agents building agents” as a likely pattern, but others note Pi is “just a harness” and conceptually simple.
- There is concern about low-quality, LLM-written GitHub issues (“slop”) that ignore templates and mix observation with speculation.
- A few propose better logging, templates, and stronger expectations around concise, human-verified bug reports.
Naming and confusion
- Several readers find the name “Pi” confusing given Raspberry Pi’s popularity and note the article is not about that hardware.