Notes and Takeaways from AI eats the world
Go here for the talk or scroll down for my notes.
Want to get my future notes when I publish them? Subscribe to my weekly newsletter below.
My notes
Slide 10 — The SaaS → LLM Shift: “What Can We Automate Next?”
Enterprises historically adopt more software as each platform shift reduces friction—mainframes (1 app) → PCs (dozens) → SaaS (400–500 apps). The new shift to LLMs asks a different question: what can now be automated, not what new app do we add?
This platform shift is not about adding more tools. It’s about converting repetitive, human-driven tasks into automated workflows.
Slide 11 — Uncertainty in Capabilities: “We Don’t Know How Good This Gets”
Previous platform shifts had predictable improvement curves. With LLMs, leaders disagree—some believe AGI-level leaps are close; others think multiple breakthroughs are still necessary.
Startups should design their AI strategy around optionality—building capabilities that benefit them today but also position themselves to quickly adopt step-change improvements in accuracy and automation.
LLM capability curves are uncertain. In the near term, startup AI adoption should emphasize modular, replaceable AI components so future model improvements compound value rather than require re-architecture.
Slide 12 — Is AI One Platform Shift or Many?
We should assume AI will create multiple waves of enablement:
Automation
Agentic workflows
Real-time multimodal intelligence
A multi-wave platform shift means startups must treat AI as a continuous capability—not a one-time initiative.
Slide 50 — The Pattern of New Tech: Absorb → Innovate → Disrupt
Every technology wave follows a pattern:
Absorb: Make existing workflows more efficient
Innovate: Build new products & bundles
Disrupt: Redefine the problem entirely
Most startups are currently in the “Absorb” phase (internal workflows, enrichment, research automation).
Slide 51 — Where AI Works First: Coding, Marketing, Support, Automation
Most successful AI use cases in the Absorb phase sit in functions where work is digital, repeatable, and text-based.
This validates focusing early investment on:
Coding
Marketing
Support
Automation
AI’s first wins come in high-volume knowledge workflows.
Slide 63 — “AI Gives You Infinite Interns”
AI dramatically lowers the cost of cognitive labor—any repetitive knowledge task can be outsourced to a near-infinite supply of automated agents.
One opportunity is to convert repeatable knowledge tasks into an ‘infinite intern’ workflow—freeing humans for judgment, strategy, and relationships.
Slide 64 — AI Still Makes Mistakes: Human in the Loop Questions
Error rates will not disappear. Key questions become:
Which errors matter?
Can we automatically verify?
When do humans need to be in the loop?
Startups must design AI systems with embedded guardrails:
Validation layers
Confidence scoring and thresholds
Human review steps
Logging/observability of agent actions
Startup AI adoption must assume imperfect outputs. Human-in-the-loop and verification layers are required for customer-facing workflows.
Slide 65 — Jevons Paradox: More Output, Not Fewer People
Automation rarely reduces total labor—it increases productivity, so teams do more.
AI will expand what each employee can accomplish:
CSMs handle 2–3× the account load with higher quality
RevOps produces 10× deeper insights
Marketing ships more campaigns and variants
AEs prospect more, but with far more precision
AI should be viewed as an accelerant. Startup teams should produce dramatically more output, not merely the same output with fewer people.
Slide 70 — Automating What We Know: 10–20× More Creative Output
AI collapses the cost of asset creation—ads, images, copy, variations—enabling organizations to create dramatically more content.
Marketing and GTM can immediately benefit:
10–20× more ad variants for testing
Automated segmentation-based email/landing pages
AI video outputs for campaigns (nonprofit + commercial)
Personalization at scale across both verticals
With generative AI, startups can test and deploy far more targeted assets across channels—reducing CAC and improving conversion rates.
Slide 86 — When Automation Works, It Disappears (Elevator Attendants Example)
When automation becomes reliable, it becomes invisible. Nobody thinks about elevator attendants today—the work has shifted to the machine.
What AI automates for startups will eventually become “just the way Startups work.”
AI should ultimately fade into the background—becoming an invisible, reliable layer that powers every workflow inside Startups.