I was hired to help with a billing-system migration at Bluehost. The job was supposed to be straightforward: JustHost and HostMonster were being acquired, and their systems needed to move into the Bluehost software ecosystem. That migration turned out to be more technical than anyone anticipated, which meant the billing job became hands-on work with WordPress, web hosting, domain registration and DNS settings — not just processing payments.
I didn't plan to learn SEO. I learned their systems fast, got promoted to team lead, and followed the technical work wherever it led. By the time I left, I'd moved into a role training a 60+ person department on SEO and website performance, and my next job was as an SEO strategist at 180Fusion. The path from billing team lead to SEO specialist doesn't exist on any org chart. It happened because a migration made the work interesting, and I stayed curious about the parts that weren't in the job description.
Twelve years later, the same pattern played out with LEGION — except this time the unplanned curriculum was a mobile creature-battler game I built for no reason other than wanting to see if I could.
The toy came first
I started the game in October and November 2024. It's a turn-based battler in Godot 4.6: players collect procedurally generated creatures called chimeras, assembled from five independent part slots with deterministic stat reconstruction, and fight them in a type-advantage battle system. The game is parked right now, waiting on one specific thing — an automated 2D-art pipeline good enough to produce the game's art at the volume it needs. The tech and game engine are fully functional and tested.
The game mattered because it was a toy: no business case, nothing riding on it.
I designed the battle math in a spreadsheet first — damage formula, agility bands, type chart, part stat budgets — then specified it precisely enough for an agent to implement and cover with headless tests. One prompt from November 2024: "You are the Product Manager for a video game called Chimeria. Your task is to create stories for the development team to follow… Ask questions until you are 95% certain you have the information you need to create a development roadmap."
That "ask questions until 95% certain" pattern reappeared in my LEGION chats five months later.
LEGION's first appearance in the record is March 2025. The game predates it. It is entirely because I was messing around with making a toy that I developed the skills necessary to build the advanced machinery I later developed: specify in prose, verify the math against a known-good source, make the agent prove it with tests, keep state in files the agent can read back.
By the time Google Cloud said "agentic orchestration has emerged as a new paradigm" in March 2025 and OpenAI launched the Agents SDK a few weeks earlier to solve "customers struggling with custom orchestration logic," I was already building LEGION around the same architectural idea — model plus tools plus loop plus state — while the industry was still naming it.
The loop closed both ways
On 20 March 2026, I ported LEGION's own automation apparatus back into the game. An unattended Planning → Execution → Refinement → Execution chained sprint sequence ran for a day and took the engine's test count from 651 to 1,725. Same day, a dedicated refinement audit found and fixed a production bug that unconditionally purged all active-tier creatures from the database on every app launch (a "DEV ONLY"-marked function that shipped live), a forced-switch bug that always gave the enemy a free turn, missing defensive-stat flags that silently diverged from the design spec, and a SQL-injection-shaped string-interpolation weakness.
The skills moved from toy to production tool, then the production tool came back and made the toy better.
Eight months of not building it
LEGION itself started as a curiosity project that failed for most of a year.
March 2025: I wanted AI APIs to talk to each other on my computer, then thought it might work better as a web app. The vision was simple to say: I wanted to be able to have a discussion with the VP about strategic initiatives while the research, problem solving, asset acquisition and fulfillment were handled by the AI organization without further input from me. The browser gave me a 404.
April 2025, attempt two: Docker plus FastAPI plus a LangChain task runner. Agents would update files within the virtual environment, which would be deleted when the session ended. I ran a test with an early AI agent where they edited the look of the UI, which resulted in black text on a black background. The agent didn't seem to believe it was capable of anything and defaulted to not reading the handbook unless it was prompted to do so.
May 2025, attempt three: LEGION v2.0 in Lovable. I was on a free account, which cannot update GitHub. I switched to a pro account. It said the build was still unsuccessful and gave me more errors.
LEGION came together after I learned to work with coding agents and new models arrived in December 2025. The first working files on disk are dated 13 March 2026. On 10 September 2026, roughly 18 months after I first described it, LEGION was actualized into a system powerful enough to rebuild the entire website within a couple of hours.
I didn't take a class on AI. I didn't read a textbook. I played with it. I used it for stuff that had nothing to do with work. I built a creature-battler with AI just because I wanted to see if I could. All of that play taught me how AI actually thinks. By the time I needed to use it for business, I already knew how to use it well.
The path from a billing migration to SEO strategist and the path from a toy game to LEGION are the same path. Something got interesting enough to follow. I followed it without asking whether it was in my job description or whether it would ever be useful. Both times, it turned out the thing I learned while I wasn't trying to learn it was exactly the thing I needed when the work got hard.