February 7, 2026
Flow vs. bot fail
Haskell for all: Beyond agentic coding
AI coding agents promised miracles — devs say they just kill the vibe
TLDR: The author argues AI coding agents hurt productivity and break a developer’s focus, pointing to studies and interviews. Commenters clap back with drama: some say agents are immature but evolving, others demand transparency and speed, and many want AI as a helpful assistant—not the one making big decisions.
The post says “agentic coding” — AI bots that try to code on their own — isn’t making developers faster or happier. The author cites personal trials, job interviews where agent-using candidates performed worse, and studies like Becker and Shen, then argues for calm, flow-friendly tools instead. But the comments turned it into a soap opera. First flare-up: “This wasn’t really about Haskell,” scoffed one reader, branding the title clickbait. Another dropped the MS‑DOS era meme, saying today’s AI is clunky but will grow up — like training wheels for coders. A heated thread blasted agent tools for being black boxes: “You zoom out, let it run, come back to a mess,” one said, demanding dashboards that show every decision. Interview drama hit next: “Using coding agents in interviews?” Critics called it nonsense without a clear written spec. Meanwhile, the chill crowd rallied behind AI as an assistant, not a boss: let it handle boring boilerplate and help with reviews, but keep humans in the driver’s seat. The speed demons blamed flow-killing waits on slow models, name-dropping ultra-fast providers like Cerebras and swearing real-time responses feel different. Verdict from the peanut gallery: agents need transparency, speed, and manners — or they’re just noisy interns
Key Points
- •The article argues current agentic coding tools often reduce productivity and developer familiarity with codebases.
- •Evidence cited includes personal experience, candidate interview outcomes, and research studies by Becker and Shen.
- •The Becker study’s screen recordings indicate developer idle time roughly doubled when using agentic coding.
- •The author’s guiding principle is to keep users in a flow state, suggesting agentic tools frequently break flow.
- •Calm technology principles are proposed, with inlay hints in IDEs (e.g., VSCode) as a non-LLM example that preserves context and flow.