Mr. Phil Games

The full field guide to how I drive Claude Code — design docs, phased delivery, verification gates, and continuous review — now lives on my personal site.

S&box launches April 2026 as an open-source engine on Source 2.

What happens when you give an AI agent autonomous control of a persistent game world — the tick architecture, guardrail system, and lessons from building Deep Hollow.

I'm entering Deep Hollow into the Solana Agent Economy Hackathon — building a tradeable item economy and AI agent trading in five days with Claude Code.

475 commits, a new Unity frontend, a partnership with Pixel Cows, and a design pivot toward shorter play sessions.

Deep Hollow is a survival-strategy game where your AI assistant defends your underground fortress while you're away.

Jevons' paradox — the 160-year-old economics principle — explains why AI tools will create more demand for developers, not fewer.

Every programming language was designed for humans.

AI is about to flood the market with more games than any storefront can sort.

Meta lost over $83 billion on the metaverse.

Everyone's talking about the multipolar world.

Robot labor at $2/hour won't kill indie games — Jevons' paradox says it'll unleash millions more.

AI is about to flood the market with games.

How I replaced my entire marketing stack with Claude Code skills, MCP servers, and terminal tools.

Block cut nearly half its workforce and Wall Street cheered.

Jai — from closed beta to the upcoming open-source release, community growth, and how it compares to Odin and Zig.

AI is to programming what the printing press was to hand-copying manuscripts.

The DoD threatened to blacklist Anthropic for refusing to remove AI safety restrictions.

MyFitnessPal acquired Cal AI for its photo-based calorie tracking.

A template letter to send your representatives about the Department of Defense's threat to designate Anthropic a 'supply chain risk' for maintaining AI safety restrictions.

Stripe's AI coding agents produce over 1,300 merged PRs per week with zero human-written code.

The Model Context Protocol loads every tool definition upfront, burning up to 30% of your AI's memory before you type a word.

Agentic coding is the biggest shift in software development since the IDE.

Your AI coding agent is brilliant but clueless about your game.

Three months of updates: Changesmith launch, first YouTube video, Stellar Throne progress, Agentic Game Devs community, and more.

Fleet navigation, event system, research persistence, and construction system all fixed and implemented. 13 commits, 1000+ lines changed.

Fixing star rendering, preserving planet data, restoring entity names, and fleet composition through Zig serialization improvements.

Building a comprehensive serialization parity test framework, fixing floating-point precision, and validating the dual-engine architecture.

The Zig simulation backend is now live in production.

Comprehensive parity testing, 11 bugs fixed, and the Zig backend reaches production-ready status with 52.7x performance improvement.

Hey everyone — MrPhil here with another deep dive into Stellar Throne.

All 21 simulation phases complete in Zig with 52.7x speed improvement.

All 23 game managers fully ported from GDScript to Zig. 333 tests passing, TOML config migration complete, and simulation system track 2 underway.

Ground combat gets strength bars, ownership updates, and battlefield grids.

A multi-day struggle with a persistent combat timing bug, exhausting every debugging strategy before a final standard approach resolved it.

Major planetary warfare upgrades with Orbital Bombardment and Ground Invasion systems, redesigned Ground Combat UI, and dozens of bug fixes.

Four core reasons for building ClaudeCraft: democratizing game creation, empowering creative communities, learning through teaching, and embracing AI's creative future.

Combat UI scroll fix in Stellar Throne and ClaudeCraft course structure progress with modules 7-10 complete.

Comprehensive combat UI screen infrastructure redesign and launching the ClaudeCraft 10-week program for AI-assisted game development.

Refining combat UI, building a Zig-based test harness, and creating specialized Claude agents for development and quality control.

Upgraded to Opus 4.1, migrated to GUT testing framework, and evaluated six combat UI layout designs.

*From low-level power to high-level polish — how a week-long experiment with Godot reshaped my AI-assisted workflow on Stellar Throne

Upgraded to Godot 4.3, performed a full codebase audit, and continued combat UI fleet layout improvements.

Newsletter about switching Stellar Throne to a new engine.

Chasing elusive bugs to bring combat back online, fixing fleet behavior, engagement triggers, and optimizing loading screens.

AI-assisted refactoring with Claude requires clear boundaries.

Fog of war restored, star field in place, animated notifications implemented, and CLAUDE.md earns a 9.5/10 audit score.

The Godot port reaches functional completion with fleet travel, 96 technologies, fog-of-war, and interactive panels all working.

Huge porting strides across combat, strategic layer, and UI systems.

Stellar Throne officially reaches Alpha.