Richard Garriott says artificial intelligence has changed the maintenance picture for Shroud of the Avatar so dramatically that the long-running MMO now reaches a bug count of “zero” every day.
The claim came during a discussion of AI’s growing role in game development, where Garriott argued that automation is helping his team address defects that could once linger for years. He described the game as having historically carried at least 1,000 known bugs at a time, including lower-priority issues that remained unresolved for three to five years. Now, he said, problems are fixed within an hour of being found.
“Shroud of the Avatar, for example, now has zero bugs, every day. Zero,” Garriott said.
That is an enormous statement for any continually updated online game, and it needs careful interpretation. It does not necessarily mean that no player can ever encounter an unexpected behavior, crash, exploit, performance problem, or newly introduced defect. Garriott did not provide a public definition of what qualifies as a bug in that count, how issues are detected, whether the total refers to confirmed and tracked tickets, or whether fixes have passed every possible form of validation. Those missing details matter as much as the headline number.
Still, the comments point to a practical use of AI that is distinct from the more divisive conversation around generative art, writing, voice work, and code produced from a text prompt. Garriott’s emphasis was on finding, prioritizing, and eliminating technical problems faster—work that sits closer to quality assurance, debugging, regression checking, and routine engineering maintenance.
From a persistent backlog to rapid fixes
A bug backlog is the accumulated list of known defects that have not yet been fixed. In an MMO, that list can include anything from an interface element behaving incorrectly to a quest condition failing, an ability calculating damage improperly, a server-side issue, or a rare problem caused by a particular sequence of player actions.
Not all reports receive the same urgency. Teams generally sort defects by severity and priority. A problem that blocks progress for many players or threatens account, economy, or server stability is likely to be treated as urgent. A small visual error or an obscure edge case may be acknowledged but deferred while developers handle more consequential work. Garriott’s description of bugs sitting for years reflects that reality: a defect can be real and documented without being the next task a limited team can justify taking on.
His central argument is that AI has changed the cost of that work. If tools can help identify an issue, trace its likely cause, suggest or prepare a fix, and assist with checks after the change, more of the backlog becomes economically possible to tackle. In Garriott’s framing, that means fewer people and less time are needed to make and maintain a game.
Related coverage includes Richard Garriott Says AI Has Taken Shroud of the Avatar's Bug Count to "Zero".
That is both the appeal and the fault line. Faster technical work can improve a live game for players, particularly when it prevents familiar issues from becoming permanent fixtures. But reducing the number of people required for development also directly connects AI adoption to fears of job displacement. Garriott acknowledged that the topic is controversial and said ethical use remains important, including questions about whether AI systems draw on other creators’ work and how voice-related data is handled.
What “zero bugs” can—and cannot—tell players
“Zero bugs” sounds absolute. In software practice, it is usually more useful to ask what exactly has reached zero. It could describe the number of open issues in a team’s current tracker after fixes are made. It could mean there are no known confirmed defects at a daily checkpoint. It could exclude duplicates, reports awaiting reproduction, minor visual defects, or issues considered design behavior rather than bugs. None of those possibilities have been specified here.
That does not make the claim meaningless; it establishes its limit. A bug tracker measures known and categorized work. Players, meanwhile, judge a game through their actual experience: whether it loads reliably, performs acceptably, lets them complete content, preserves progress, and responds consistently to their actions. Those two views can align, but they are not identical.
There is also a timing issue in any game that receives regular updates. A newly added feature, content change, balance adjustment, or technical revision can create new problems even after a previous batch has been cleared. In that sense, daily movement toward zero is more plausible as a workflow claim than as a permanent declaration that future defects cannot exist. Garriott specifically connected AI to fixing bugs as soon as they are found, which describes an ongoing maintenance cycle rather than a final endpoint.
For players, the most meaningful potential benefit is therefore not a slogan but a turnaround time: if a legitimate issue is discovered, reproduced, corrected, and safely deployed much faster than before, the game can spend less time in a visibly broken state. The important unanswered question is whether that speed holds up across all kinds of issues, especially the difficult ones that only appear under unusual player behavior or broader live-service conditions.
AI-assisted debugging is not the same as handing development to a machine
Garriott characterized the division of labor in relatively direct terms: people make the art, while machines can handle work that ensures the details are checked so the game is stable, fast, and enjoyable. That framing positions AI as assistance for systematic labor rather than a substitute for every creative decision.
In technical terms, debugging is the process of locating the cause of a software defect and changing the relevant code or data so the unwanted behavior no longer occurs. The change itself is only one stage. Developers also need to make sure the fix solves the original issue without damaging something else. That second concern is often called regression testing: checking that a repair has not caused a previously working feature to fail.
AI can theoretically be useful at several points in that pipeline. It may help review reports, identify patterns among recurring issues, search a large codebase, summarize logs, propose tests, or flag changes with potential side effects. But the supplied comments do not identify which tools are being used, how independently they operate, what human review remains in place, or how the team verifies a recommended fix. Those distinctions are essential when evaluating any assertion that AI has delivered unusually fast results.
The same caution applies to “bug detection.” A system that identifies potential trouble is not automatically proving there is a reproducible defect. Reports need to be triaged: developers must determine whether an event is a genuine bug, intended behavior, incomplete information, a duplicate, or a problem tied to conditions that cannot yet be recreated. An efficient system could substantially reduce that workload, but it does not remove the importance of judgment.
The tension behind the productivity pitch
Garriott’s comments capture a wider argument now facing game development. On one side is the possibility of cheaper, faster iteration. A studio that spends less time buried in repetitive maintenance may be able to keep a game stable, react to reports more quickly, and direct more attention toward creative work. For a project with a large historic defect backlog, that could be a material improvement.
On the other side is the question of who benefits when fewer people and less time are needed. Development jobs are not just a budget line; they involve specialized knowledge, creative judgment, testing experience, and accountability. The concern is especially sharp when AI is presented as a route to reduce staffing rather than simply support people already doing the work.
Garriott did not claim to have solved the ethical questions around AI. He explicitly raised concerns involving possible plagiarism and voice training, while saying his team is attempting to use the technology ethically. That acknowledgement is important, but it is not the same as a detailed policy. The conversation leaves open practical questions about data provenance, consent, oversight, credit, and how much decision-making remains with human developers.
For live games, there is another reason to separate promise from proof: maintenance is never only a matter of closing tickets. Players need confidence that reports are heard, that serious issues are addressed, and that each update is not creating a fresh cycle of trouble. Communities built around persistent online games are familiar with long-term upkeep as a challenge; even marathon achievements in other MMOs, such as this 6,572-hour Old School RuneScape pet speedrun, underline how much time players may invest in a constantly maintained world.
A striking claim, awaiting measurable detail
Shroud of the Avatar has been available for eight years and continues to receive updates. That combination makes rapid bug resolution especially relevant: maintenance is not a one-time finish line when new work can introduce new errors.
Garriott’s assertion is notable because it describes a dramatic reversal from a game that, by his account, long operated with a four-figure bug count. Yet “zero” is most responsibly read as a report about the team’s present issue-management process, not as independently demonstrated proof of a flawless MMO. Without a published methodology, a visible tracker, definitions for included issues, or detail on validation, outsiders cannot determine the scope of the figure.
The underlying claim remains consequential even with those caveats. If AI assistance truly allows a small team to find and resolve known defects within hours rather than leaving them in a multiyear queue, it could offer a meaningful model for supporting aging live games. Whether that model earns player confidence will depend less on an absolute number and more on observable stability, transparent standards, and the human responsibility behind every automated step.
