<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Salvi Haus Studio devlog</title><description>Salvi Haus Studio is an AI-native game studio: one developer, four browser games, and agent-driven workflows that punch above their weight. Devlogs, case studies, and a games portal.</description><link>https://salviha.us/</link><language>en-us</language><item><title>Salvi Haus: one developer, four games, and a lot of agents</title><link>https://salviha.us/log/salvi-haus-one-developer-four-games/</link><guid isPermaLink="true">https://salviha.us/log/salvi-haus-one-developer-four-games/</guid><description>How a two-person studio keeps four browser games in development: AI agents do most of the labor, and the humans direct and verify it.</description><pubDate>Tue, 11 Aug 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Salvi Haus Studio is two people. One of us writes code and the other runs the business. We have four browser games in development at once, all written in TypeScript and distributed through Discord and the open web, with no installs and no outside funding.&lt;/p&gt;
&lt;p&gt;A one-developer studio with four games in flight sounds implausible, so it deserves an explanation: AI agents do most of the labor, under the developer&apos;s direction. &quot;AI-powered studio&quot; is a phrase that earns suspicion, so this post describes what the arrangement means in practice.&lt;/p&gt;
&lt;h2&gt;The division of labor&lt;/h2&gt;
&lt;p&gt;The split: humans do judgment work, agents do located, verifiable work.&lt;/p&gt;
&lt;p&gt;Judgment work is deciding what the game is — which mechanic is the core loop, whether a feature makes the game better or just bigger, whether the thing that got built last night is fun, what ships next. None of that is delegated. An agent will optimize whatever you point it at, so a human has to choose the target.&lt;/p&gt;
&lt;p&gt;Located, verifiable work is everything where you can say &quot;the task is here, and here&apos;s how we&apos;ll know it&apos;s done&quot;: implement this system against this interface, fix this bug with a failing repro, write content that fits this schema, wire this UI to this state. Games are full of this kind of work. It accounts for most of the hours in game development, and it&apos;s the part that used to make a four-game studio impossible for one person.&lt;/p&gt;
&lt;p&gt;Verifiability is the requirement that shapes everything else. If agent output can&apos;t be checked by tests, types, a schema, or playing the build, then delegating it creates work instead of saving it. A lot of our engineering effort goes into making tasks checkable so they can be delegated: typed interfaces between systems, content formats with validation, worlds defined as data.&lt;/p&gt;
&lt;h2&gt;Four games is a portfolio, on purpose&lt;/h2&gt;
&lt;p&gt;We run four games at once deliberately:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;&lt;a href=&quot;/games/dacha/&quot;&gt;Dacha!&lt;/a&gt;&lt;/strong&gt; — a cozy town for your Discord server; every guild gets a persistent farming/town world on real-world time. Our flagship, in playable alpha, headed for the Discord App Directory.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;&lt;a href=&quot;/games/annexi/&quot;&gt;Annexi&lt;/a&gt;&lt;/strong&gt; — instant browser territory strategy, with real-time matches that run 15 to 20 minutes. Mid-sprint toward a free web launch.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;&lt;a href=&quot;/games/adventure-forth/&quot;&gt;Adventure Forth!&lt;/a&gt;&lt;/strong&gt; — a persistent pocket-MMO for a few dozen friends, with a world built by agents. Playable today, early as a product.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;&lt;a href=&quot;/games/dum-runner/&quot;&gt;DÛM RUNNER&lt;/a&gt;&lt;/strong&gt; — browser co-op extraction shooter. A playable vertical slice, next in the queue.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The games share a technology spine and a distribution channel (Discord plus the open web), so tooling and lessons carry over between them. They&apos;re also at different stages on purpose, which means the studio always has something to ship and something to learn from. The status labels on the game pages mean what they say.&lt;/p&gt;
&lt;h2&gt;What this devlog is&lt;/h2&gt;
&lt;p&gt;This site is our only marketing channel. We&apos;ll write about development progress, design decisions, and playtest findings, and about agent-driven development itself, since that&apos;s the most interesting thing we know about. Alongside the devlogs we&apos;ll publish case studies of the workflows: how tasks get specified, how output gets verified, and what fails. If you want to know how two people keep four games moving, the case studies will show the machinery.&lt;/p&gt;
&lt;p&gt;The devlog has an &lt;a href=&quot;/rss.xml&quot;&gt;RSS feed&lt;/a&gt;, and playtests for all four games happen in the studio Discord. If any of this sounds useful to you, both are open.&lt;/p&gt;
</content:encoded><category>studio</category></item></channel></rss>