Summary: AI changes the economics of enterprise software: by doing tasks humans did rather than helping them work, it can deliver ~10× SaaS value and justify contracts of millions per year. That makes playbooks newly viable: a top-down enterprise motion in small non-US markets, country by country, exploiting the adoption lag outside the US. Forward-deployed engineers install and productize, and the ambition is to become an enterprise’s single, horizontal AI partner, an “AI operating system”, instead of one point solution among many. Usage-based AI spend clashes with how enterprises budget, so pricing that gives every employee room to try AI (per-user credit buckets) supports the AI transformation customers are buying.
Key ideas
- Value thesis: SaaS makes people more efficient; AI performs tasks done millions of times by thousands of people, so contract values reach millions (ignore-instructions—24-wonderful-road-to-100m, roy-lazar).
- Adoption-lag arbitrage: tech is adopted first in the US and years later elsewhere. Accelerating adoption abroad creates value you can partly capture (ignore-instructions—24-wonderful-road-to-100m).
- Unit economics flip: SaaS contracts (5M, even Greece’s ~25 large enterprises make a “unicorn per country” possible. Israel first, then 30 countries (ignore-instructions—24-wonderful-road-to-100m).
- “Local play × enterprise play”, plus Wolt’s playbook (smaller markets incumbents ignore) (ignore-instructions—24-wonderful-road-to-100m).
- Demand pull: boards see consumers using ChatGPT and push CEOs, giving product-market fit “before founding”. ~90% market demand, ~10% execution, by his telling (ignore-instructions—24-wonderful-road-to-100m).
- Use cases: voice/chat customer support (e.g. a Maccabi health-fund agent that books appointments and escalates), mortgage processing that cuts days to minutes, telco next-best-offer, warehouse/shipping/support orchestration (ignore-instructions—24-wonderful-road-to-100m).
- Forward-deployed engineers vs consulting: you’re paid for the product; each deployment’s learnings feed a generic product, so the 8th bank benefits from the first 7 without sharing their IP. Hence SaaS-like valuation multiples. Client engineers build most agents on the platform (ignore-instructions—24-wonderful-road-to-100m).
- Horizontal over point solutions: Wonderful avoids being a “customer-support vendor” in RFPs and sells full AI transformation. Vision: one AI per enterprise owning whole functions (support, back office, HR, legal), and an AI OS per vertical (~11 verticals) in five years (ignore-instructions—24-wonderful-road-to-100m).
- Outside the US, geography matters more than industry: a bank and a telco in Austria behave more alike than two banks in different countries (ignore-instructions—24-wonderful-road-to-100m).
- North-star metric: must-win accounts across countries, not near-term revenue; initial contracts can expand 10–30× (ignore-instructions—24-wonderful-road-to-100m).
- Enterprises plan budgets in advance and struggle with usage-based AI spend. Today they have experimentation budgets; long term they’ll have to plan usage (presentations, images, videos) or buy all-in solutions at a premium until they learn the real cost (startup-for-startup—369-ai-pricing, roy-mann)
- Pricing for AI transformation: per-user credit buckets let every employee try AI without one power user exhausting the org’s credits, and whoever builds a production agent moves to an uncapped pool (startup-for-startup—369-ai-pricing, roy-mann)
Disagreements & open questions
- Horizontal AI partner vs vertical point solutions: Wonderful argues buying separate AI for support, back office, and HR is a mistake; vertical vendors (e.g. support-only agent companies) compete on one slice (ignore-instructions—24-wonderful-road-to-100m).
Takeaways
- When evaluating AI vendors, weigh one horizontal platform against stacking point solutions per function (ignore-instructions—24-wonderful-road-to-100m)
Related
second-brain · ai-rd-rollout · agent-workspaces · voice-agents · ai-moats · ai-pricing