← back
Where AI is superhuman: The right jobs to automate with LLMs
Takeaway
AI is superhuman where volume kills humans and only rules-based legacy software competes — target SOC, customer engagement, supply chain ops.
Summary
- Andy Triedman (Theory Ventures partner) shares research on where LLM automation creates startup/investment opportunities, based on hundreds of buyer/builder interviews.
- Three LLM emergent properties driving workflow automation: transformation (format conversion, the historical blocker), synthesis (e.g., Gemini/OpenAI Deep Research summarizing hundreds of sites), and reasoning (modeling written-reasoning distribution — strong on Stack Overflow / code / math, weaker on tacit domain expertise).
- Three-tier opportunity spectrum: complex low-volume jobs (strategic planning) → co-pilot; mid-complexity (40-70% automatable) → end-to-end workflows with humans as reviewers; high-volume low-complexity → LLMs are already superhuman because they outcompete legacy rules-based software.
- Drop Zone AI: agentic SOC investigations replacing rules-based alert triage where analysts only see ~1% of alerts and quit in 12-18 months (4M-analyst global shortage).
- Amp (customer engagement): replaces rules-based marketing journeys with experimental agentic personalization across millions of users — marketers become experimentalists crafting variants.
automationai-businessjobs
Original description
Ask an LLM system to do any job and it'll give it a try. But not all jobs are made equal – the nature of different work means that LLMs will be much more disruptive in some areas compared to others. If you're a founder deciding where to build, or an executive deciding where to invest, what should you do? In this talk we dive into first principles assessment of what LLMs do best, what types of jobs see the best technology-workflow fit, and which jobs LLMs will disrupt entirely because they're already superhuman in capabilities.