VP PM's Deep Dive: 4 Real Pain Points of Coding with LLMs and the Human Edge
Greedy_Rise_6567 · reddit · 2026-07-31
A VP-level PM with 16 years of experience shares insights from months of heavily using LLMs (like Claude) for a side project. While AI massively boosts efficiency, he argues humans still hold decisive edges in four areas:
- Memory & Context: LLMs have a short memory. Over a project's lifecycle, developers morph into "AI project managers," forced to constantly feed context via Markdown docs to prevent the AI from spinning out of control and generating high-quality garbage.
- Compounding Hallucinations: In long coding sessions, even minor hallucinations incur huge costs. He had to build a coherence harness and use adversarial reviews to keep the AI honest.
- Expertise & Taste: AI tends to provide generic, run-of-the-mill solutions. Bridging the gap between "good" and "great" still relies heavily on human taste and domain expertise.
- Originality: AI explores possibilities quickly but mostly in a straight line. Humans, however, make weird intuitive leaps to stumble onto something genuinely original.
In summary, he views AI as the best intern he’s ever had, but not yet the best architect.
Related event: Practitioners Reflect on LLM Capabilities and Boundaries(2 posts)→
More from AGI Musings
- A missing parameter in scaling laws: H, the human effort behind learning signals — lateinteraction · 2026-10-16
- First-ever paper on multi-agent swarm safety lands NeurIPS spotlight — LauraRuis · 2026-10-16
- Yoav Artzi on peer review: paper flooding still pays off as review system strains under load — yoavartzi · 2026-10-16
- Letta founder: work in fields defined by problems, not methods — lateinteraction · 2026-10-16
- Why pure mathematicians will soon flood into AI x-risk debates — Dr_Atoosa · 2026-10-16
- RSI began when model outputs augmented next-gen training data, agents just escalated it — akbirthko · 2026-10-16