TL;DR
A programmer describes being skeptical of large language models for agentic coding while still finding them useful as a 'digital clerk' for searches and reference. He argues that some vocal LLM proponents pressure others and suggests that intense evangelism may reflect insecurity rather than purely objective gains.
What happened
In a personal commentary, a developer who identifies as skeptical of LLM-driven productivity lays out his mixed experience. He uses LLMs mainly to fetch documentation, search the web, and occasionally for constrained coding tasks when given a small context and firm constraints. By contrast, experiments with agentic systems — what he calls prompt-driven development or "vibe coding" — were frustrating: the models required extensive babysitting, produced small and often incorrect edits, and drained tokens without producing reliable results. He does not object to others using these techniques, acknowledges they enable non-experts to build projects, and notes he sometimes gets paid to clean up such work. The author criticizes aggressive proponents who frame skeptics as fearful of change and proposes that some evangelism may stem from projection and insecurity. He also leaves room to be proven wrong and invites reflection from advocates.
Why it matters
- Claims about LLM-driven leaps in developer productivity may not match every user's real-world experience.
- Aggressive advocacy can alienate experienced practitioners and polarize technical discourse.
- Widespread use by non-developers raises questions about code quality and follow-up maintenance.
- If evangelism is rooted in personal insecurity, it complicates conversations about adoption and skill gaps.
Key facts
- The author identifies as an LLM productivity skeptic.
- He finds LLMs useful for searching the web, finding documentation, and looking up algorithms.
- He uses LLMs for limited coding when the context is small and guidelines are clear.
- Agentic LLM approaches (prompt-driven development or "vibe coding") were disappointing in his trials.
- Those agentic workflows required a lot of 'babysitting' and produced small, often incorrect code changes.
- The author felt less capable as tokens were consumed during agentic sessions.
- He acknowledges these tools enable many non-experts to create things they otherwise couldn't, and he sometimes gets paid to clean up that work.
- He criticizes some vocal proponents for characterizing skeptics as resistant due to fear of obsolescence.
- The author hypothesizes that intense evangelism may be projection born of insecurity.
- He remains open to changing his view and invites LLM advocates to consider their own skill levels.
What to watch next
- Not confirmed in the source: whether agentic LLM workflows will mature into consistently reliable tools for experienced developers.
- Not confirmed in the source: whether prominent evangelists will acknowledge limitations or adapt their messaging to avoid personal attacks on skeptics.
Quick glossary
- Large language model (LLM): A neural network trained on large amounts of text to generate or transform language-based outputs in response to prompts.
- Agentic LLM: An LLM configured to take multi-step actions or manage tasks autonomously, often via chains of prompts or tooling integrations.
- Prompt-driven development: A workflow where developers rely on prompts to guide an LLM to produce code or make changes, rather than writing the code themselves directly.
- Vibe coding: An informal term for using generative models to create or evolve code with an emphasis on iterative, conversational prompts.
- Tokens: Discrete units of input or output text used by LLMs; many models and pricing schemes meter usage by token count.
Reader FAQ
Does the author use LLMs at all?
Yes — he finds them useful for web searches, finding documentation, and limited coding tasks with small context and clear instructions.
Did agentic LLMs work well in his experience?
No — he reports they required heavy supervision, made many small incorrect edits, and were slow and frustrating.
Does he oppose others using LLMs?
No — he accepts that these tools enable people without formal development training to build projects and says others should use what works for them.
Why does he criticize LLM evangelists?
He objects to vocal proponents who describe skeptics as fearful of change and suggests that some evangelism may be driven by insecurity or projection.
Is there evidence about placebo effects or perceived gains?
The author notes, in a footnote, that some evidence suggests a placebo effect, but he does not elaborate; further details are not provided in the source.
1/14/2026 The Insecure Evangelism of LLM Maximalists I am an LLM productivity skeptic. I find LLMs useful as a sort of digital clerk – searching the web for me, finding…
Sources
- The insecure evangelism of LLM maximalists
- My AI skeptic friends are all nuts
- 5176031.pdf
- Roland Tanglao
Related posts
- Inside the White House shitposting machine: AI-altered political clips
- Why the Yarn Spinner Team Refuses to Use AI in Product and Development
- The Insecure Evangelism of LLM Maximalists and Its Impact on Developers