Here is my honest take after watching the AI space for the first four months of 2026.
The hype-to-delivery ratio has improved in some areas and gotten dramatically worse in others.
What is actually delivering:
- AI coding tools — GitHub Copilot, Cursor, Claude for Code are producing real productivity gains with real adoption curves
- Multimodal pipelines — Natively multimodal models are now the baseline, not the exception
- Enterprise automation — AI agents completing multi-step workflows at 70–80% reliability in controlled environments
- AI search and research — Perplexity-style answers are genuinely useful for knowledge workers
What is still mostly hype:
- AGI claims — every “major milestone” announced in Q1 was followed by a correction within weeks
- AI agents in uncontrolled environments — failure rate outside demos is still unacceptable for most production deployments
- AI hardware breakthroughs — chip-level claims are real but enterprise impact is still 12–18 months out
- AI replacing meaningful white-collar jobs — net job impact is a wash; displacement is concentrated in very specific task types, not entire roles
The pattern: AI is genuinely good at tasks that are well-defined, high-volume, and tolerant of occasional errors. Anything requiring judgment, context-switching, or relationship management remains human territory for the foreseeable future.
Most expensive AI failures I am seeing come from treating category-2 problems like category-1 ones.
What is your experience been? Push back if you disagree — I actually read the comments.
Opinions based on ongoing industry tracking. I could be wrong.
