Krishna Kumar Sharma builds Omukai, a Berlin sovereign AI startup for Europe that cannot be cut off by the US or China.
Omukai's single interface commands five robots and five drones by natural-language voice, with basic tasks today and hundreds within two years.
His book Agentic Coding: The Boring Way says cross-review by Claude Opus, GLM 5.2 and Codex beats token maxing for ROI.
Sovereign AI lets Europe deploy and control its own systems
Krishna Kumar Sharma left Amazon to build Omukai, a Berlin startup selling sovereign AI that Europe can deploy and control itself, beyond the reach of US or Chinese sanctions. The same interface commands five robots and five drones by voice, and Sharma says teaching a robot one task now takes a week rather than years. His book argues agentic coding shares AI's wider problem, where burning tokens substitutes for earning returns, and he expects prices to jump when subsidies end.
Takeaways
Sovereign AI must be self-deployed. Sharma defines sovereign AI as a system no other country controls and that you can deploy yourself. His Berlin startup Omukai builds AI in Europe for Europe, so neither the US nor China can sanction or cut off its services.
One interface commands robots and drones. Omukai lets one user command five robots and five drones by natural-language voice. Basic tasks work today, and Sharma expects complex jobs like folding clothes within two years, when a robot will understand hundreds of tasks instead of being taught one at a time.
Robot training now takes weeks, not years. A simple task that once took years to teach now takes a week or two, because robots learn from videos and special devices. Sharma says world models let a robot attempt tasks it has never seen, completing an uncomplicated one within a day.
Defense is where AI stakes are highest. Sharma says AI is treated like nuclear technology because it scales destruction, cyber warfare and propaganda. Drone swarms identify, track or attack targets and intercept missiles, and recent wars used hundreds of thousands of drones. Europe's guardrail leaves the decision to attack with a human.
Token maxing burns budget without return. His book, Agentic Coding: The Boring Way, attacks burning tokens on simple tasks. He defines productivity as $100 or $200 of benefit from $10 spent, cites Gartner research that heavy AI use costs more than human work, and says layoffs mainly enriched Anthropic and OpenAI.
AI costs will rise when subsidies end. He predicts costs rise four to five times once subsidies end, and that users already struggling for productivity will stop paying. He notes roughly 98% of AI spending goes to coding and productivity, while defense is about 2% of total defense budgets.
How the cross-review method works
Use at least three models from different labs. Sharma's set includes Claude Opus, Codex from OpenAI and China's GLM 5.2.
Have each model review the others' work instead of trusting one answer.
Roughly 50% of the findings differ, so the disagreements expose gaps a single model would miss.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
The models correct each other and fill in the gaps, which Sharma says reaches Fable 5-level quality at a lower price.