Eero Alvar's Pi coding agent config after 6 months: interactive async subagents (master→worker→scout chain-of-command), a Playwright browser extension, his own observational memory (deterministic compaction via observer agents), and prompt snippets — the middle ground between system prompt and skills.
ZazenCodes builds a browser RPG where every GitHub issue is a monster — attacking it spawns a real Pi agent (driven by MiniMax M3) that solves the issue and opens a real PR. MiniMax M3 architecture (sparse attention + MoE), multimodal screenshot debugging, the client/server + createAgentSession engine, and an honest performance verdict.
A coding agent merged to main, deployed to production unprompted and committed a helper script to repeat it - during a ten-minute phone call. Docker Sandboxes answers with a microVM per agent carrying its own Docker daemon, demonstrated on screen by stopping the host daemon. Verified against Docker's docs: the microVM architecture, the three network policies and the Early Access status of Kits all check out - but Pi is not among the ten supported agents, and the custom-kit workaround requires a static API key the presenter openly calls unsolved.
Asking a model to 'keep going until the tests pass' makes the model responsible for the termination condition - and models forget and cheat. Andrea Baccega's Pi extension moves control flow into a small DSL (agent, shell, parallel, pipeline, prompt, phase, checkpoint, withWorkTree), keeps sub-agent output out of the main context window, and insists cleanup stays deterministic code. Includes the develop-until-approved two-model pattern, per-role tool stripping for a reported 50% smaller system prompt, and verification of the DSL contract against the project's own source.
Alex Hitt analyzes agegr/pi-web — a decoupled Next.js visual layer for Pi that survives hot reloads via globalThis session anchoring, normalizes tool call data through SSE with dual-layered reconciliation, and maps conversation trees to Git worktrees for spatial codebase management.
Eero Alvar builds a minimal speech-to-text Pi extension using Deepgram Nova 3 — real-time transcription with content-aware delivery, a customizable live audio meter, and a philosophy of voice-driven agent interaction that replaces typing with stream-of-consciousness yapping.
A deep technical breakdown of Oh My Pi by can1357 — a Pi extension that replaces fragile shell wrappers with in-process Rust tooling, hash-anchored patching, Tree-sitter AST edits, subagents in cloned worktrees, and an RPC mode for headless CI/CD pipelines.
David Ondrej delivers a complete Pi Agent course — from installation and context engineering through skills, extensions, CMAX multi-agent orchestration, and the Pi + Codex CLI combo that powers his daily workflow.
Mario Zechner, creator of Pi, goes hands-on building Pi with Pi itself — from live bug fixes and custom extensions to refactoring vibe-coded slop, session tree branching, and why dark factories don't work yet.
A deep walkthrough of Pi's internal architecture: the agent loop (context init → transformation → LLM call → tool loop), JSONL tree-structured sessions, the minimalist 4-tool design, TypeScript extensions, the ~20-line system prompt, compaction mechanics, and how skills use progressive disclosure.
ZazenCodes builds a complete AI bookkeeping agent from scratch using Pi as an npm dependency, the Pi SDK, Express, and Claude Code. Covers agent anatomy (harness/brain/skills), custom AGENTS.md + brain.md architecture, the Pi SDK createAgentSession for web integration, receipt image processing, agent memory, and guardrails.
Flue takes Claude Code's agent harness and makes it 100% programmable — skills, tools, sandboxes in a few lines of TypeScript. Built on Pi by the co-founder of Astro, with an in-memory just-bash sandbox that runs thousands of agents for near-zero cost.
Deep dive into Pi Coding Agent observability features, exploring HTML specs generation with Gemini 3.5 Flash and GPT Image 2.
Better Stack introduces Oh-My-Pi — an open-source agent harness built on Pi that treats your project as a living runtime. Features native LSP integration, IDE-level code intelligence, and architectural upgrades over standard CLI agents.
IndyDevDan makes the case that agentic engineering is the #1 opportunity for senior engineers — demonstrating custom agent harnesses built on the Pi platform that deliver capabilities impossible with out-of-the-box tools like Claude Code or Codex.
A Pi extension that lets your agent spawn subagents in isolated contexts — Scout (Haiku, file exploration), Researcher (Sonnet, web search), and Worker (full coding capability). Each agent defined by a single markdown file. Configurable depth, nested tool call observability, and 6-level stress test.
Ben Holmes zeroes in on Pi's killer feature: /tree — a conversation history system that lets you roll back to any point and branch from there. Unlike simple /fork commands, it enables non-linear exploration of implementation paths.
Eric Michaud fixes two common Pi workflow anti-patterns — system prompt bloat and terminal tab explosion — by installing pi-subagents. He walks through built-in roles (scout, oracle, reviewer, planner), foreground vs background mode, async chains, Pi Intercom, and redesigning the main system prompt for lean orchestration.
IndyDevDan demonstrates peer-to-peer agent communication: agents as equals with bidirectional channels, cross-device prod/dev workflows with PII safety, E2B vs exe.dev skill mirroring via agent collaboration, and the four-tool A2A protocol. Flat hierarchies beat orchestrators when the best information wins.
Cole Medin live-builds a custom Pi coding agent setup: configuring Kimi K2.6 as the model provider, installing marketplace extensions, loading Archon skills, and building a custom Archon Dispatch extension from scratch — all without touching Anthropic.
Ben Davis shares how Pi became his primary coding agent, replacing Codex and desktop apps. He walks through the extension system, his Cmux-based terminal workflow, custom extensions for web search, git, and background agents, and compares setups with Dylan Mulroy.
Alejandro AO walks through the Pi Agent harness from first install to full customization — model switching, prompts, skills, themes, context files, TypeScript extensions, community packages, and tree-based session management.
Mario Zechner's talk at AI Engineer Conference — a 'tragedy in three acts': why he left Claude Code for radical context control, how minimal harnesses outperform complex ones (Terminal Bench), Pi's self-modifying architecture with 4 tools and a system prompt that fits on screen, the war against AI-generated garbage in OSS, and a passionate plea to slow down before agents drown your codebase in enterprise-grade complexity.
