TL;DR
A developer describes skepticism about using large language models (LLMs) for agentic, prompt-driven coding, finding them useful for searching and limited tasks but disappointing for autonomous programming. The author argues vocal proponents may be projecting insecurity and invites reconsideration of who benefits from current LLM workflows.
What happened
In a blog post dated January 14, 2026, an author labeled themselves an LLM productivity skeptic while outlining how they use these models. They find LLMs helpful as a 'digital clerk' — for searching documentation, looking up algorithms, and occasional constrained coding tasks with clear context. By contrast, the author reports a poor experience with 'agentic' LLMs and prompt-driven development: implementations required heavy supervision, delivered slow and often incorrect small code changes, and left the user feeling less competent as tokens were consumed. The post criticizes insistently optimistic advocates who portray LLMs as an inevitable revolution and attribute skeptics' resistance to fear of obsolescence. The author suggests that some evangelism may stem from projection and insecurity, while also admitting they could be wrong and that learning to use agents might be a distinct skill. Footnotes mention some evidence of placebo effects and the author's availability for hire.
Why it matters
- The debate shapes developer expectations about what LLMs can reliably do in production coding.
- Overstated claims from advocates risk polarizing communities and mislabeling legitimate skepticism as fear or resistance.
- If agentic workflows require substantial human oversight, promised productivity gains may not materialize for all users.
- Perceptions of LLM capability influence who adopts these tools and how teams allocate time to training versus cleanup.
Key facts
- The author identifies as skeptical about LLM-driven productivity.
- They use LLMs for tasks like web searches, finding documentation, and constrained coding with small contexts.
- Their experience with agentic, prompt-driven development was disappointing and required a lot of babysitting.
- Reported issues included slow, small, and often incorrect code changes and a feeling of diminished competence as tokens were consumed.
- The author observes vocal proponents framing LLM adoption as inevitable and diagnosing skeptics as emotionally resistant.
- They argue some evangelism may be projection of insecurity by users who see agents as superior.
- The author acknowledges the possibility they are wrong and that mastering agent usage could be a distinct skill.
- Footnotes in the source mention limited evidence suggesting a placebo effect and state the author is available for hire.
What to watch next
- Whether agentic LLMs become less error-prone and require less human supervision over time – not confirmed in the source.
- How developer communities respond to differing narratives about LLM productivity and whether framing shifts – not confirmed in the source.
- If empirical studies clarify productivity impacts of prompt-driven development versus traditional workflows – not confirmed in the source.
Quick glossary
- LLM (Large Language Model): A machine learning model trained on large amounts of text to generate or predict language and assist with tasks like drafting, summarizing, or coding.
- Agentic LLM: An LLM configured to act autonomously on multi-step tasks, often executing actions, invoking tools, or managing state with limited human intervention.
- Prompt-driven development: A workflow where developers rely on prompts to an LLM to generate code, tests, or other software artifacts rather than writing them directly.
- Vibe coding: An informal term for using LLMs interactively to produce code based on high-level directions or 'vibes' rather than detailed specifications.
- Placebo effect: An apparent improvement in performance or perception stemming from belief or expectation rather than a direct causal effect of the tool or treatment.
Reader FAQ
Does the author find LLMs completely useless?
No. They report LLMs are useful for searching the web, finding documentation, and limited coding with small, clear contexts.
Did the author try agentic, prompt-driven development?
Yes — they describe trying it and finding the experience disappointing, citing heavy supervision and frequent errors.
Is the claim that evangelists are insecure presented as fact?
The post presents that as the author's interpretation and critique, not an independently verified fact.
Are there definitive studies proving agentic LLMs increase productivity?
Not confirmed in the source.
Is the author available for hire?
Yes; the post's footnote states the author is available for hire.
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
- Hacker News
- The Transformative Influence of LLMs on Software …
- 5176031.pdf
Related posts
- Google Translate testing alternate translation options to boost accuracy
- Reality Check: Generative AI Falling Short on Trust, Value and Scaling
- ListenBrainz restricts APIs after AI scrapers overload MusicBrainz services