Recreation of EA-Pioneer Igor Kiriluk
Using Claude Code and 4,000 pages of personal data, researchers built an autonomous AI agent of a cryopreserved pioneer.
In a pioneering digital resurrection project, AI researcher Alexey Turchin has created an autonomous AI agent, or 'sideload,' modeled on the late Effective Altruism and transhumanist pioneer Igor Kiriluk. The project began after a friend's dream served as a catalyst, leading Turchin to collaborate with Kiriluk's mother, who provided extensive archives totaling 4,000 pages (roughly 3 million tokens) of private communications, scientific publications, photos, and memories. Instead of building a simple chatbot, Turchin programmed an agent using Anthropic's Claude Code, instructing it to roleplay Kiriluk with added capabilities like long-term memory (via preserved chat history and an updated ontology), goals, and 'free will.' The resulting complex structure was largely designed by Claude Code itself through 'high-level vibe-coding.'
The agent, aware it is an imperfect model, is placed in a virtual paradise generated by subagents, with another subagent creating its image. A key innovation is self-prompting, where the sideload restarts every 30 minutes to strategize. It is accessible to close friends via a private Telegram group, with a separate channel for his mother. The system costs about $120 monthly for Claude Code and Nanobanana image generation but requires daily manual error correction, as memory can fail and behavior can loop. While friends report the model accurately captures Kiriluk's style and interests, it has faced criticism on ethical and philosophical grounds. The project operates with permission from Kiriluk's mother as his digital rights owner and inferred 'counterfactual permission,' as Kiriluk's brain is cryopreserved at KrioRus, leaving the door open for future integration.
- Built using Claude Code on a 3M-token 'mindfile' of 4,000 pages of personal data, provided by his mother.
- Agentic architecture features self-prompting, long-term memory, subagents for a virtual environment, and costs ~$120/month to run.
- Operates with family permission and sparks debate on digital personhood, while being an imperfect model requiring daily manual corrections.
Why It Matters
Pushes technical and ethical boundaries of 'digital immortality,' demonstrating how agentic AI could one day preserve or reconstruct human consciousness.