CTO and Co-Founder of Poolside, building AI agents that can reason about and modify complex codebases at scale. Previously founded source{d} (AI for code) and Athenian (data-enabled engineering platform). Co-hosts the Developing Leadership podcast with former GitHub CTO Jason Warner.
Building the Path to AGI Through Software Development
Eiso Kant represents a pragmatic approach to AGI: prove AI capability through increasingly complex, verifiable engineering tasks. His thesis is that software development is the ideal test domain for AGI—it provides clear success metrics, verifiable outcomes, and real-world complexity that captures the full spectrum of human problem-solving.
Current Work at Poolside
At Poolside, Eiso is building AI agents that go beyond code generation to achieve genuine architectural understanding. The company has raised $623M to build what Eiso calls a “model factory”—a production-grade system designed for iteration speed over scale. Key achievements:
- Researchers can push new RL optimization ideas to production in under a week
- Architecture sweeps of 500+ experiments run over a weekend
- 20% of training corpus is synthetic data (targeting 98% in coming years)
- Models trained as agents in reinforcement loops using real codebases
Rather than treating code as tokens to predict, Poolside’s approach focuses on understanding design intent, navigating massive codebases, and making autonomous modifications while maintaining architectural integrity.
Background & Previous Ventures
source{d} (2015-2019) - Founded the world’s first company dedicated to applying AI to code and software engineering at scale.
Athenian - Built a data-enabled engineering platform focused on helping teams understand and improve their development workflows.
Tyba (2011-2016) - Co-founded a recruitment platform connecting developers with companies.
Philosophy: The AGI Lens
Eiso frames Poolside’s work within the broader AGI narrative. Software development serves as an ideal proving ground because:
- Clear Success Metrics - Code works or it doesn’t; tests pass or fail
- Verifiable Outcomes - Automated testing provides ground truth
- Incremental Difficulty - Scales from simple bug fixes to architectural redesigns
- Production Constraints - Real engineering captures requirements benchmarks can’t measure
“If you want to compete at the frontier of model capabilities, the secret isn’t just scale—it’s iteration speed.”
Key Insights & Contributions
Model Quality Over Scaffolding - Competitive advantage comes from models sophisticated enough to reason about real codebases, not elaborate engineering tricks.
Synthetic Data as Foundation - Embracing model-generated training data refined through RL, guided by small layers of high-quality human feedback.
Agent-Centric Architecture - The model is the agent, operating inside reinforcement loops using real-world codebases as training environments.
No Scaling Wall - In his Air Street Press interview, Eiso argues synthetic data generation and search techniques will enable continued progress toward AGI.
Notable Appearances & Writing
- RAAIS 2025 - “Inside poolside’s path to AGI” - Candid walkthrough of Poolside’s model factory and AGI strategy
- World Economic Forum - Agenda Contributor on AI and software development
- Developing Leadership Podcast - Co-host with Jason Warner, discussing engineering leadership lessons
- Air Street Press - Featured in “There is no scaling wall” discussion on AI scaling and AGI progress
Conference Appearance
Event: AI Engineering Code Summit 2025 Date: November 21, 2025 Time: 4:00 PM - 4:19 PM Session: “AGI: The Path Forward”
Eiso discussed Poolside’s vision for achieving AGI-level capabilities in knowledge work, how software development serves as a proving ground for AGI, and the engineering challenges in building truly autonomous coding systems. His presentation emphasized moving from code generation to code reasoning, from suggestions to autonomous action, and from autocomplete to architecture-aware modification at production scale.