Companies are still misreading LLM workflows, and ROI is hard to prove
teropa · x · 2026-07-25
The thread argues that companies are still getting AI adoption wrong: natural-language interfaces feel intuitive, but LLMs are not humans and have very different strengths and failure modes.
- Simply giving everyone access and asking them to “figure it out” has not produced reliable workflows.
- Many companies are spending heavily but cannot measure outcomes.
- Where ROI is measurable, some are seeing engineering spend rise while output only improves modestly, with the poster citing a 10–30% curve lift as the kind of result that is already surprising.
- The core point: natural language is a deceptively easy interface, but productive AI use still needs deliberate workflow design, not just broad access.
More from Companies & People
- NVIDIA CEO Meets South Korea’s President to Deepen AI Partnership — nvidia · 2026-07-25
- Vercel COO Uses AI Agents to Scale 10-Person SDR Team Down to Just 1 — baseten · 2026-07-25
- KOL Accuses Anthropic of Distilling Competitor's Model to Build Opus 5, Calls for Sanctions — 0xAllen_ · 2026-07-25
- Prentis, New AI Lab by Reid Hoffman and Marc Pincus, Seeks $100M — TechCrunch AI · 2026-07-25
- A16z survey names Elad Gil, Marc Andreessen and Elon Musk as top angels — jfischoff · 2026-07-25
- The Washington Post runs Ask The Post AI on independent inference with 1.79B tokens a month — togethercompute · 2026-07-25