← All speakers

Bio, Work & Ideas

Mario Zechner

Conference affiliation: PI · 2026

On this page

Mario Zechner created libGDX and pi, open-source tools spanning two generations of software development: cross-platform games and AI-assisted programming. At Earendil, he leads pi’s technical direction, building coding agents that developers can inspect, customize, and control.

From game development to coding agents

Zechner created libGDX, a Java game-development framework supporting desktop, mobile, and browser platforms. Its users included Ingress, Slay the Spire, and Spine, the animation software he also worked on. He originated the scene2d architecture underlying libGDX’s user-interface tools and largely transferred stewardship to the framework’s community by 2016.

He subsequently worked on RoboVM, which enabled Java applications to run on iOS, and built its first proprietary add-on: a debugger. Xamarin acquired RoboVM, closed its core, and was itself acquired by Microsoft, which shut the project down. Community contributors forked the available code, restored missing capabilities, and preserved libGDX’s iOS support through MobiVM. The experience made the practical right to fork central to Zechner’s approach to open-source governance.

Zechner built pi after finding coding assistants increasingly opaque: system prompts changed unexpectedly, tools altered model context without adequate visibility, and customization remained constrained. Its minimal architecture separates model-provider abstraction, the agent loop, terminal rendering, and the coding interface. Four default tools—read, write, edit, and shell execution—provide the foundation for TypeScript extensions that can add commands, providers, interfaces, security policies, and context-management strategies.

These self-modifying agent extensions can be generated by pi itself using bundled documentation and examples, then reloaded immediately during an active session. Zechner distributes additional capabilities through pi-skills, which also supports Claude Code and Codex CLI.

Pi reached a wider audience after Peter Steinberger incorporated it into OpenClaw’s agentic core. The resulting influx of automated, low-quality contributions prompted Zechner to require new contributors to open concise, human-written issues before submitting pull requests. He also clusters related reports and periodically closes issue trackers to protect maintainers’ time.

In April 2026, Earendil acquired pi and Zechner joined the company. He became a shareholder and retained responsibility for pi’s technical direction alongside founders Armin Ronacher and Colin Sidoti. Pi’s core remains MIT-licensed and forkable; future commercial additions may use fair-source or proprietary terms.

Principles shaping his work

  • User-controlled model context. Developers should understand what information reaches their models and avoid hidden prompt changes, premature diagnostics, and unexplained context pruning.
  • Extensibility without workflow lock-in. Subagents, planning modes, integrations, and security controls should remain optional, composable extensions instead of mandatory complexity.
  • Open-source maintainer protection. Automated submission volume is not community health; human-written explanations, contribution filters, and temporary tracker closures defend maintainers’ attention.

Talks by Mario Zechner

1 talk

Key ideas

Scroll to read ↓

Mario Zechner explains how dissatisfaction with opaque coding-agent harnesses led him to build pi, a minimal, extensible agent, and why sustainable AI-assisted development depends on bounded tasks, deliberate review, and human ownership.

  • Own the context: Hidden prompt changes, injected reminders, premature diagnostics, and pruned tool outputs can change agent behavior in ways developers cannot easily inspect or control. 1:24 ↗ 2:25 ↗ 3:16 ↗
  • Keep the core small: Pi combines a short system prompt, a simple tool-calling loop, and four built-in tools: read, write, edit, and bash. 5:13 ↗ 6:21 ↗
  • Make customization part of the workflow: TypeScript extensions, bundled documentation, broad hooks, and hot reloading allow users and agents to adapt the harness during an active session. 6:21 ↗ 7:21 ↗ 8:16 ↗ 9:16 ↗
  • Protect maintainer attention: Human-first contribution gates, issue clustering, and temporary tracker closures can help counter the volume of low-quality automated open-source contributions. 10:10 ↗ 11:11 ↗
  • Bound agent work to what humans can verify: Modular tasks, clear evaluation criteria, and limits on generated code reduce the risk that local agent decisions accumulate into system-wide complexity. 13:19 ↗ 15:24 ↗ 16:31 ↗
  • Keep critical judgment human: Read critical code, make consequential design decisions yourself, and treat the friction of implementation as a source of understanding rather than a problem to eliminate. 16:31 ↗ 17:44 ↗

References