Customer service is the killer enterprise AI use case today
Customer service offers quantifiable ROI (deflection rates), natural escalation paths for risk mitigation, and existing infrastructure — making it the fastest enterprise adoption vector.
AI agents will eat human labor spectrum from both ends
Agents will simultaneously augment highest-paid knowledge workers (coding) and replace lowest-cost repetitive labor (customer service), driven by ROI clarity at both extremes.
AI agents become the universal brand front-end replacing apps/websites
Conversational agents will unify sales, support, and commerce into a single persistent interface with full user context, becoming the primary brand touchpoint.
Voice-to-voice is the next frontier but hallucination rates 8x higher
Voice is the natural human UI, but voice-to-voice models currently suffer ~8x higher hallucination rates than text, requiring hybrid architectures for enterprise deployment.
Conversation data flywheels create compounding moats for agent companies
LLMs can ingest 100% of customer interactions to auto-generate evals, detect gaps, and improve agents continuously — creating a data advantage that compounds with tenure.
Application layer captures most value; model APIs face commoditization
Solving the end-user business problem allows application companies to command pricing power, while model providers face low switching costs and intense competition.
Fine-tuning small open models for specific tasks beats monolithic large models
Mature applications can decompose agent workflows and fine-tune smaller models for routing, classification, and guardrails — improving latency, cost, and reliability.