When Your AI Won't Parallelize: Debugging Invisible Model Behaviors
How I discovered that my planning agent was creating fully sequential task chains despite explicit instructions to maximize parallelism — and why switching models fixed it.
Jon Kloss
Software engineer building AI tooling, agents, and workflows — plus indie games and developer tools along the way.
About
About Me
Hey, I’m Jon — a software engineer with about a decade of experience building across the full stack. By day I ship production software; by night I’m usually tinkering on side projects just because they sound fun.
Right now I’m especially drawn to AI tooling and indie games. I love making tools that help other developers move faster, and I’ll pick up whatever language or framework the problem calls for.
Outside of code, you’ll find me camping, gaming, or hanging out with my dog.
Experience
Where I’ve Worked
Aug 2023 – Present
United Wholesale Mortgage
Full-timeSoftware Developer IIIJul 2025 – Present
Software Developer IIAug 2023 – Jul 2025
Jun 2022 – Jun 2023
bash.gg
Full-timeSoftware Engineer IIJun 2022 – Jun 2023
May 2020 – Jun 2022
Merkle
Full-timeSoftware EngineerMay 2020 – Jun 2022
Sep 2017 – May 2020
HelloWorld, Inc.
Full-timeAssociate Software EngineerSep 2017 – May 2020
Education
Wayne State University2012 – 2017
Skills
Rust
TypeScript
C#
Python
GDScript
C
React
Next.js
React Native
Tauri
Node.js
.NET
PostgreSQL
Redis
Unity
Godot
Claude Code
Docker
Git
Projects
Things I’ve Built
Open Source
claude-workflowwireWire — a fast, local-first API client built with Tauri, Rust, and ReactCheckmate-ChaosanvilnephsAI Creative Workshop for Kids — Claude Code custom skills that let kids build games, stories, and charactersdiablo-loot-helperDiablo 4 Loot Analyzer — evaluate drops against your build with AI-powered scoringFlecs.NETA C# wrapper for flecsclaude-code-skillsClaude Code skills for GitHub PR review and responseagent-testTesting framework for AI agents. Record, replay, assert, and detect regressions.
Blog
Writing & Thinking
The Bloated Checklist Problem: When More Instructions Make AI Worse
My AI agent was hitting 50-iteration loops on simple tasks. The fix wasn't better prompting — it was less prompting.
Building a Semantic Cache for LLM Responses
How I built a three-stage caching system that serves repeated queries in microseconds instead of waiting seconds for API responses — with git-aware invalidation.
Teaching AI to Self-Correct: A Six-Stage Verification Pipeline
How I built a graduated verification system that catches AI coding mistakes at the cheapest possible stage — from instant syntax checks to multi-reviewer consensus.
Beyond Text Diffs: Building a Structural JSON Comparison Engine
How I built a path-aware JSON diff engine for API regression testing that understands arrays, nested objects, and type changes — not just string differences.
Writing a Chess AI That Understands Pieces It's Never Seen
How I built a minimax AI for a chess roguelike with 21 piece types — including a Necromancer that raises undead pawns and a Bomber that explodes on death.
Contact
Get in Touch
Have a project in mind or just want to say hello? I’d love to hear from you.