← sources

As an engineer, coding is the least interesting part: running a live MMO solo with Claude Code

source: reddit (reddit:ClaudeCode)published 2026-10-29 04:48ingested 2026-10-30 02:40

retained text

I'm a solo dev building [**Whispered Scrolls**](https://whisperedscrolls.com), an action combat fantasy MMO. You can level twenty skills from 1 to 99 and fight with real time aimed combat, no tab targeting. Parties consist of real tank, damage and healer roles. There are no classes, and you can change how you play at a moments notice. There are 20+ dungeons, world bosses, 160+ quests, a player market and several PvP modes. You can jump in within seconds from a browser on desktop or phone. Dedicated apps run on Windows, Linux and Android. Every version additionally comes with full controller support. The game has spent a few weeks in open alpha. Our players love it, and an ever growing group of people log in and play every day.

My background spans SaaS, networking and also solo game development. These years have taught me one thing above all. Coding is the least interesting part of an engineer's job. It's a lot tedium, a means of expressing systems. What truly matters is when building a game like this is the infrastructure, systems, and gameplay design, plus the tooling that validates and certifies the work. This takes real human effort to get right. Claude Code took the tedium off my plate. Now most of my days go into design, playtesting, and validation. I wanted to share what that looks like in practice

# Design is the job

These days I spend more time on infrastructure, systems and game design than on code. Think skill curves, how gathering feeds crafting and what each gear tier is worth.

A lot of work comes down to saying no. Players often ask me to add one thing or change another to suit their taste. This doesn't mean I am not open to feedback. While I've been able to create a much better experience directly from user feedback, plenty of those ideas clash with the game's vision, its balance or how characters progress. Claude will cheerfully build anything you ask for. That makes knowing what *not* to build an incredibly important skill.

# Validating the design, not just the code

* **Simulations.** I simulate progression before any number changes. Thousands of runs cover skilling methods, XP rates and gear paths. They show how long things take and where players outrun the content. For example, players once told me levelling felt too fast. So I ran the analysis against a production backup. Per-hit XP matched the sim exactly. However, the curve itself was too generous and the content ran dry around level 75. An argument about feel became a quantifiable decision about numbers.
* **Tests that certify a release.** The full suite runs once per release and gets recorded against the main git tree hash. If the code hasn't gone green, the deploy flatly refuses to ship it.
* **Network and performance checks.** A lag and packet-loss proxy recreates bad connections on demand. One rubber-banding report turned out to be TCP head-of-line blocking. You only find that kind of gremlin if you know where to look. Performance changes also ship with before-and-after benchmark numbers.

# The open alpha is key to success

No sim catches everything, no matter how many runs you throw at it. Open alpha's biggest gift is exposing design failures I never thought to model. That's why the whole setup exists to close the loop quickly.

* **In-game bug reports.** Players press a button and describe what went wrong. Along with their words, I get a screenshot plus their character, world and position. The report also carries a client state snapshot, the client version, recent server events and hardware details. Most bugs reproduce without a single follow-up question. Claude triages them through an admin CLI and I pick what goes in the next batch. Reports close once the fix goes live. One recent batch cleared 47 minor issues.
* **Analytics built around questions.** The admin panel asks plain questions. Are people staying, is the story landing and is progression paced right? Is gold worth anything and where do people get stuck? Each number is a rate per player or per hour, with medians and p90s. I split them by level band and account age. Every chart marks each deploy. Checking whether a patch worked takes a single glance.
* **Live tuning.** I can change nearly every balance number while the game runs. That covers mob health and damage, ability cooldowns and radii, drop tables, shop prices and gathering speeds. Each change reaches every connected client so tooltips stay accurate. The server logs who changed what. Once the numbers feel right, one command exports them into the repo. The next deploy then bakes them in. When players find a hole, I can close it the same day which buys time to design a better solution moving forward.

# Shipping like it's SaaS (because it is)

One command handles the whole deploy. It starts a five-minute in-game countdown, holds new logins and stops the game. Then it takes an offline database backup, runs migrations, restarts and reopens logins. If anything fails, it restores the backup and rolls back to the previous image. It also pings a staff Discord channel so I know right away. The game simply won't start while a migration is pending. Afterwards, patch notes post to Discord by themselves. The script also updates the RSS feed and the main site.

The desktop apps are thin shells around the live game. As a result, one deploy reaches web and desktop players at the same moment. Overall, a server deployment takes 30 seconds and is seamless for players.

# Keeping parallel agents and usage under control

* **Contracts before code.** An architecture doc and a canonical ID list carve the game into owned areas. Those areas are shared simulation, server, client and content. Each agent works in its own git worktree, inside its own area.
* **One queue for heavy work.** Typecheck, tests, builds and screenshots all wait in one machine-wide queue.

 This avoids waste for competing machine resources, which is a real issue even with a powerful machine.

* **Usage as a budget.** Each role gets the right model and each agent gets a short brief instead of the whole design doc. Agents also stop and ask me the moment they need a decision. Instead of wasting tokens on assumptions or burning expensive tokens for minute mechanical changes, changes are focused, unclear outcomes refused, and spend is optimized.

# Gamedev is hard, and MMOs are never finished

Indie MMOs wrestle with the same problems AAA studios do. Big worlds have thousands of interacting systems and endless ways for characters to express themselves. Nobody gets that right on the first try. Every major MMO fights balance problems. Every one of them also has players slipping outside the systems built for play. MMOs are never finished. That's exactly why they're **soft**ware and not **hard**ware. Software is malleable and meant to bend, change and keep improving.

What Claude Code changed for me is how fast I can iterate. Ensuring both technical and gameplay systems are scalable, maintainable, and lead to a quality end product are key things Claude cannot own, as does the validation that a change is right.

Want to see the result? It's free to play at [**whisperedscrolls.com**](https://whisperedscrolls.com/), in a browser on PC or phone. Desktop apps cover Windows and Linux (Steam Deck included). There's an Android app as well. The whole game also has optional controller support. A Steam release is upcoming, after the open alpha wraps.. Come say hi on Discord at **discord.gg/ERJmmVutFa**. I'm happy to dig into any part of the setup in the comments.

document fetched 2026-10-30 00:43 · canonical https://reddit.com/r/ClaudeCode/comments/1xiaq02/as_an_engineer_coding_is_the_least_interesting