A game developer at a workstation overlooking a colorful coastal landscape with hikers.

Big Walk AI: What House House Actually Used It For

Key Takeaways

  • Big Walk is a cooperative multiplayer adventure for 2 to 12 players, developed by the Melbourne studio House House and published by Panic. It launched on 4 August 2026 for PC, Mac, PlayStation 5 and Nintendo Switch 2 at $19.99.
  • On 9 November 2026, 19 jurors for the 2027 Independent Games Festival said they would decline to vote for any game that uses generative AI in any capacity.
  • On 10 November, House House added a section to the Big Walk website's FAQ disclosing that large language models were used "sparingly" in the final stretch of development, to help with difficult technical challenges and bugs.
  • The studio estimates that "well under 0.1%" of the game's code had anything to do with that use, and says generative AI was not used in any other context when making the game.
  • The main example given is a rendering bug in Unity's Universal Render Pipeline that made the game run about 50% slower for around 20 seconds at sunrise and sunset. The fix, after extensive modification of a chatbot's suggestion, came to about 10 lines of code.
  • House House says it made no disclosure on Steam because Valve's rules do not request disclosure of code or chatbot use, and admits the game was developed without an internal AI policy — a lack it calls "a failure".
  • Steam reviews turned sharply negative after the disclosure: on 10 November, only 36% of that day's reviews were positive, with 156 negative reviews against a single negative review the day before.

What Big Walk is

Big Walk is a cooperative multiplayer adventure about exploring, solving puzzles and finding ways to communicate with friends. That description, used by the publisher Panic, captures the game more accurately than any genre label. There is no combat. There is no single-player mode. The game is designed to be played start to finish with a group of friends, in a strange open world, talking to each other as they walk.

The numbers describe the shape of it. Big Walk supports 2 to 12 players, with crossplay between PC, Mac, PlayStation 5 and Nintendo Switch 2, so friends on different platforms can join the same walk. It launched on 4 August 2026 at $19.99, with a 25% launch discount on some storefronts. It is built around proximity voice chat: players can only hear each other when their characters are close together in the game world, which means the social experience — shouting across a valley, losing track of each other, finding each other again — is the game, as much as the puzzles are. The puzzles themselves are environmental and cooperative: groups solve them together to open coloured towers, unlock the black tower, and eventually walk to the black sphere on the horizon.

The game comes from House House, a small Melbourne studio best known for Untitled Goose Game, the 2019 hit in which players control a goose whose job is to annoy the residents of a village. That game was a cultural phenomenon, and it set the expectations for Big Walk: a small team, a distinctive idea, and a game that trusts players to find the fun in each other's company rather than in combat or competition. Big Walk is the studio's follow-up, and by the measures that matter to a publisher, it has been a considerable success. Panic announced on 10 August 2026 that the game had passed one million copies sold in six days, quoting the line "In just six days, over one million of you have bought Big Walk". The Steam page shows a Metacritic score of 91 for the PC version, and the PlayStation 5 version was a PlayStation Plus Essential monthly game from its launch in August 2026. Reviewers have largely praised it as one of the best co-operative releases of the year, with some calling it the opposite of the cheap, disposable multiplayer games that dominate the market.

That success is part of why the events of November 2026 matter. Big Walk is not a marginal game, or a game whose reputation was already in doubt. It is one of the best-reviewed and best-selling independent games of the year, made by a studio with a strong reputation, and it is now at the centre of the most heated argument in games about generative AI. The argument is not about whether the game is good. It is about whether a game can be good and still be disqualified — from awards, from goodwill, from the trust of its players — because of how a small part of it was made.

The week everything changed

The controversy unfolded over three days, and the order of events matters, because each step changed the meaning of the one before it.

On 9 November 2026, a group of 19 jurors for the 2027 Independent Games Festival published a statement on Bluesky, shared by writer and juror Austin Walker. The jurors said there had been a significant rise in submitted games made with generative AI tools, and that it had sparked intense disagreement across the judge and jury pools. As a result, the signatories said they were "declining to vote for any game that uses generative AI in any capacity". The signatories included Walker, Sam Barlow, Lena Raine and Brendon Chung, among others. The statement did not name any games. It also did not represent the IGF as a whole: the festival's rules do not currently mention generative AI, and Walker has said there are around 50 jurors in total, so the 19 signatories are a substantial minority, not a majority. The group has asked the IGF to adopt a strong anti-AI policy for future years that would disallow games containing any AI-generated work. IGF finalists are due to be announced in January, with the awards ceremony on 3 March 2027.

The statement landed in a context that made it explosive. According to reporting by Well Played, juror Kyle LeClair had written that the most-discussed game during voting "involved almost entirely the use of AI". Bluesky users pointed to Big Walk as the game in question, though it was never officially named, and House House contests that characterisation. Whether or not Big Walk was the game LeClair meant, the speculation connected the jurors' statement to the year's biggest indie hit within hours, and the connection has shaped everything that followed.

On 10 November, House House confirmed its AI use to the games publication Aftermath, and then published a fuller disclosure in the FAQ on the Big Walk website. The FAQ section is headed "Was AI used to make Big Walk?" and it gives the studio's account in its own words. That account is the central document in this story, and it is worth reading in full rather than in summary, because the precise wording is what the argument is about.

On the same day, the reaction on Steam was swift and severe. Well Played reported that Big Walk's Steam reviews on 10 November were 36% positive, with 156 negative reviews that day, after a single negative review on 9 November. GamesRadar reported refunds and negative Steam reviews following the disclosure. A game that had sold a million copies in six days and held a Metacritic score of 91 was, within a day, being review-bombed by its own players.

What House House actually disclosed

Here is what the studio said, in substance, in its FAQ disclosure. The full text has been widely quoted, and the quotations in this article are taken from that disclosure as reported by Nintendo Life, GamesRadar and other outlets.

"In the final stretch of Big Walk's development, LLMs were used sparingly by House House to help work through difficult technical challenges and bugs. In some cases, snippets from LLM responses were copied as a starting point to be modified into game code — we'd estimate that well under 0.1% of Big Walk's code had anything to do with this usage, and generative AI was not used in any other context when making the game."

That paragraph contains the whole of the studio's claim, and each clause does work. "In the final stretch" limits the use to the end of development, not the whole project. "Sparingly" and "to help work through difficult technical challenges and bugs" limit it to a specific purpose: problem-solving, not creation. "Well under 0.1%" puts a number on it — a very small number. And "generative AI was not used in any other context" is a blanket denial that covers everything else: no AI-generated art, no AI-generated music, no AI-generated writing, no AI-generated level design. On the studio's account, the only generative AI in Big Walk is a small amount of code assistance at the end of development.

The disclosure goes on to explain what the primary use actually was. "The primary use case for House House's use of LLMs was to research, via chatbots, issues we encountered with third-party code." That is a narrower claim than it might sound. The studio is saying that the main thing it did with chatbots was not to generate code, but to research problems — to ask a chatbot about a bug in someone else's code, and to use the answer as a starting point for its own investigation. Researching a problem through a chatbot produces no code at all, which means the 0.1% figure, whatever it measures, does not capture the primary use. That is a gap in the disclosure that critics have been quick to point out, and it is worth being honest about: the studio has put a number on the code, but not on the research, and the two are different things.

The studio gives one concrete example. The team needed to cull shadow casters by distance from the camera, but Unity's Universal Render Pipeline — the rendering system built into the Unity game engine — only allows culling shadow casters by distance in the camera's forward direction. Shadow casters behind the player were being rendered at an infinite distance, which made the game run about 50% slower for around 20 seconds at sunrise and sunset. The team could not solve the problem itself, and could not solve it after asking an engineer who works for Unity. After a few dead ends, a chatbot suggested a possible fix. The team modified the suggestion extensively, and it worked. The fix, in the studio's words, "represented about 10 lines of code inside URP" — that is, inside Unity's render pipeline, not inside Big Walk's own code.

That example is doing a lot of work in the disclosure. It shows a specific, technical, unglamorous problem. It shows the team trying other routes first. It shows the chatbot's contribution being modified extensively rather than pasted in. And it shows the scale: about 10 lines of code, in a third-party rendering system, fixing a performance bug that affected the game for 20 seconds twice a day. For the studio, the example is evidence that the AI use was minimal, practical and far removed from the creative heart of the game. For critics, the example is evidence of something else — that even a respected studio, working on a beloved game, reached for a chatbot when it got stuck, and that the habit is spreading faster than the industry is willing to admit.

The disclosure ends with three statements that are as important as the facts. "This is not an endorsement of, or argument for, the use of AI tools, however minimal." That is the studio distancing itself from the technology it used, and it is a sentence that reads as both a disclaimer and an apology. "We support the work of those who are trying to limit the harm generative AI is doing to videogames and our world, and apologise that Big Walk's LLM usage may not meet their standards." That is an apology, offered in advance, to the people most likely to object. And: "Big Walk was developed without an internal AI usage policy in place, and we consider that lack of policy a failure. We will work to have a more comprehensive policy going forwards." That is an admission that the studio had no rules for a question it now recognises as important, and a promise to make rules for the future.

Why the disclosure came now

One of the first questions readers ask about the disclosure is why it arrived in November, three months after the game launched in August. The answer, on the studio's own account, is the IGF.

House House says a version of its statement went to the IGF as part of its submission, where disclosure was required under the festival's rules. The submission information and competition rules for the IGF do not mention AI in general, but the submission process asked entrants to disclose their use of generative AI, and House House complied. The FAQ on the game's website was then updated, as the studio puts it, "as it has become a matter of public interest". Reporting by Kotaku, cited in later coverage, dates the addition of the FAQ section to sometime after 23 October 2026, based on archival snapshots of the page.

That timeline matters because it answers the most damaging possible reading of the events: that the studio hid its AI use and was caught. On the available evidence, that reading is not supported. The studio disclosed its AI use to the IGF when asked, as part of a formal submission process, before the jurors' statement and before the public controversy. It then published the same disclosure on its own website. What it did not do is disclose on Steam — and the reason it gives for that is specific. House House says it has made no disclosure on Steam "as the platform does not request disclosure of code or chatbot usage". That is a claim about Valve's rules, and it is worth examining on its own terms, because it is the point on which the studio's position is most vulnerable — and most defensible.

Steam's disclosure rules and the gap in them

Steam, the PC storefront where Big Walk sells, does require developers to disclose the use of generative AI in their games. The requirement is part of the content survey that developers complete when submitting a game. But the requirement, as Valve has framed it, is concerned with AI-generated content in the game itself — art, text, audio and other assets that a player will see or hear. It is not, on the studio's account and on Valve's own documentation, concerned with the use of AI tools to write code, or with the use of chatbots to research a technical problem.

Valve's own Steamworks documentation acknowledges the AI-powered tools built into modern development environments, and tells developers that efficiency gains from those tools are "not the focus of this section". That is the gap. A studio can use a chatbot to debug a rendering pipeline, or an AI assistant to write boilerplate code, without triggering the disclosure requirement, because the requirement is about what is in the game, not about how the game was made. House House's position is that it followed the rules as written: it disclosed where disclosure was requested — to the IGF — and did not disclose where it was not requested — on Steam.

Critics have two answers to that. The first is that following the letter of the rules is not the same as being transparent, and that a studio that knew its players cared about AI use should have disclosed voluntarily, whatever the rules said. The second is that the rules themselves are the problem, and that Valve should require disclosure of all AI use, including code and chatbot use, so that players can make informed decisions. Both answers have force. The first is a judgement about the studio's choices. The second is an argument about platform policy, and it is one that Valve will now have to engage with, whether it wants to or not.

It is also worth noting what the IGF jurors' statement says about code specifically, because it closes the gap that Steam leaves open. The jurors' statement directly addresses the coding question, arguing that AI-generated code is not an exception. Their position is that "any game that uses generative AI in any capacity" should not receive their vote — and "any capacity" explicitly includes code. That is a deliberate choice, and it is the choice that puts Big Walk in the firing line. If the jurors had limited their objection to AI-generated art, music or writing, Big Walk would not be affected, because the studio says none of those were used. By including code, the jurors made the smallest possible AI use disqualifying — and that is precisely the point of contention.

The question underneath: does 0.1% count?

Strip away the timelines and the platform rules, and the argument about Big Walk comes down to a single question: does it matter that a tiny fraction of a game's code was touched by a chatbot? The two sides give opposite answers, and both answers are coherent, which is why the argument is so heated.

The studio's answer is that it does not matter, or matters very little. On its account, the AI use was minimal, late, technical and heavily modified by humans. The game is hand-crafted. Its art, its music, its writing, its level design and its puzzles are all the work of people. The chatbot helped fix a rendering bug in a third-party system. To treat that as equivalent to generating a game's art or writing with AI, the studio implies, is to confuse a tool with a creator, and to punish a team for using the same kind of assistance that programmers have used for decades in the form of search engines, documentation and colleagues.

The jurors' answer is that it matters completely, and that the size of the use is irrelevant. Their statement says they are declining to vote for any game that uses generative AI "in any capacity", and they give their reasons: the technology's widely documented negative impacts on the economy, the labour market, the environment and the global informational pipeline. For them, the question is not how much AI was used, but whether the line is drawn at all. If a line is drawn at 0.1%, they argue, it will be drawn at 1% next year, and at 10% the year after. The only defensible line, on this view, is zero — and the only way to hold it is to refuse to reward any use, however small.

That disagreement is not really about Big Walk. It is about whether generative AI in games is a spectrum or a binary — whether some uses are acceptable and others are not, or whether the technology is unacceptable in itself. The spectrum view is the one most of the industry has implicitly adopted: AI tools for debugging are fine, AI tools for generating art are contested, AI tools that replace human artists are not. The binary view is the one the jurors have now stated publicly: any use, in any capacity, is disqualifying. The Big Walk controversy is the first time those two views have collided in public over a specific, beloved game, and the collision is what makes the story important beyond the game itself.

There is a further complication that neither side has fully addressed. The studio's figure — "well under 0.1% of Big Walk's code" — does not say what the percentage is a percentage of. Lines of code, files, commits and changed lines are four different measures, and they would give four different answers. And as noted above, the primary use the studio describes — researching third-party code problems through chatbots — produces no code at all, so it cannot appear in a code percentage. The figure is therefore not reproducible, even in principle. That does not mean it is wrong. It means it is an estimate offered in good faith by a studio trying to convey scale, not a measurement that can be checked. Readers should treat it as an indication of magnitude — very small — rather than as a precise figure.

The Steam backlash, and what it tells us

The reaction on Steam was immediate, and it is worth describing precisely, because it is easy to overstate or understate what happened. Well Played reported that on 10 November 2026, the day the disclosure became public, Big Walk's Steam reviews were 36% positive, with 156 negative reviews that day, compared with a single negative review on 9 November. That is a dramatic swing, and it is the kind of swing that is usually described as review-bombing: a coordinated or semi-coordinated flood of negative reviews in response to a specific event, rather than a gradual change in opinion about the game's quality.

Two things should be said about that. The first is that the reviews were not, for the most part, about the game. They were about the disclosure. Players who had enjoyed Big Walk for three months were not suddenly discovering that it was a bad game. They were expressing anger that something they valued had, in their view, been compromised — or that they had not been told. The second is that review-bombing is a blunt instrument. It punishes the whole studio for a decision that a small number of people made, and it does so in a way that is visible to future buyers but invisible in its reasoning. A prospective buyer who sees a sudden drop in a game's review score may not know why it happened, and may simply decide not to buy a game they would have enjoyed.

Whether the backlash will last is an open question. Review-bombing campaigns often fade, and a game's long-term reputation usually reflects the quality of the game more than the controversies around it. But the damage to trust is harder to repair than the damage to a review score. House House has apologised, and has promised a more comprehensive AI policy for future projects. Whether that is enough to satisfy the players who left negative reviews — and the jurors who have said they will not vote for the game — remains to be seen.

What this means for other developers

The Big Walk controversy is a warning to every studio that has used a chatbot, an AI coding assistant or any other generative tool during development, however briefly. The lesson is not that such use is wrong. It is that the rules about disclosure are unclear, the expectations of players and jurors are stricter than the rules, and the gap between the two is where reputations are lost.

The most immediate consequence is likely to be a wave of AI policies. House House has said it will develop a more comprehensive policy for future projects, and other studios will follow — if only because they now know that the question "was AI used to make this game?" can be asked at any time, by anyone, and that the answer needs to be ready. A policy does not have to ban AI use. It has to define it: what tools are permitted, for what purposes, with what disclosure, and who decides. The absence of such a policy, which House House now calls "a failure", is the part of the story that other studios can most easily learn from.

The second consequence concerns awards. The IGF jurors' statement is not a rule change, and it does not bind the festival. But it is a public statement by 19 named jurors, and it will shape how submissions are judged, both this year and in future years. Other festivals and awards will face the same question, and they will have to decide whether to adopt a policy, and if so, where to draw the line. The jurors have urged the IGF to adopt a strong anti-AI policy for future years that disallows games containing any AI-generated work. Whether the festival does so, and how it defines "AI-generated work", will be one of the most consequential decisions in the independent games world in 2027.

The third consequence is for platforms. Valve's disclosure rules were written for a world in which the question was whether a game's art or text was AI-generated. That world has changed. The question is now whether any part of a game's production involved AI, including code, and the rules do not cover it. Valve can leave the rules as they are, and accept that players will learn about AI use from controversies rather than from store pages. Or it can extend the rules, and accept the burden of defining and enforcing them. Either choice has costs, and the Big Walk controversy has made the choice unavoidable.

For readers, the practical advice is simpler. If you are buying a game and AI use matters to you, the store page may not tell you everything you want to know. The developer's website, its FAQ, and its public statements are the places to look — and the absence of a statement is not evidence of absence. If you are a developer, the lesson is the opposite: say what you did, before someone else says it for you. House House disclosed to the IGF and then on its own website, and it still faced a backlash. A studio that discloses nothing, and is discovered later, will face a worse one.

How Big Walk compares to other AI controversies in games

Big Walk is not the first game to face a backlash over AI, and the comparisons are instructive, because they show how the argument has shifted. In the case of Arc Raiders, covered in our report on the Arc Raiders AI voice backlash, the issue was AI-generated voices — synthetic performances that replaced human actors, in a game where players could hear the difference. The backlash there was about the audible presence of AI in the game itself, and the developer's response was to re-record the voices with human performers. In the case of Nexus Mods, covered in our report on the Nexus Mods AI rules, the issue was disclosure: the platform made disclosure of AI-generated mods mandatory, but did not ban them, drawing a line between transparency and prohibition.

Big Walk is different from both. The AI is not audible, visible or readable in the game. It is in the process, not the product. A player who did not read the news would never know it was there. That is what makes the case so divisive. For those who believe AI use is a matter of principle, the invisibility of the use makes it worse, not better: it means players were not given the chance to decide. For those who believe AI use is a matter of degree, the invisibility is evidence of its triviality: if you cannot see, hear or feel the difference, the difference does not matter.

The comparison also shows how fast the norms are moving. Two years ago, the argument was about whether AI-generated art should be allowed in games at all. Last year, it was about disclosure. This year, it is about whether a chatbot used to fix a bug counts as AI use at all, and whether a game should lose award votes because of it. The direction of travel is clear: the definition of AI use is expanding, the expectation of disclosure is rising, and the tolerance for undisclosed use is falling. Studios that assume the norms of last year will still apply next year are making a mistake.

What We Could Not Verify

  • The exact scale of the AI use in Big Walk. House House's estimate of "well under 0.1%" does not state what the percentage is a percentage of — lines, files, commits or changed lines — and the primary use it describes, researching problems through chatbots, produces no code, so it cannot be captured by a code percentage.
  • Whether Big Walk was the game juror Kyle LeClair referred to when he wrote that the most-discussed game during voting "involved almost entirely the use of AI". Bluesky users pointed to Big Walk, but the game was never officially named, and House House contests the characterisation.
  • The precise timing of the FAQ update. Kotaku dates the addition of the AI section to sometime after 23 October 2026, based on archival snapshots, but no exact timestamp has been published.
  • The sales figure of one million copies in six days. That figure comes from Panic, the publisher, as reported by Game Informer. It has not been independently audited, and no breakdown by platform has been published.
  • The Metacritic score of 91. That figure is shown on the game's Steam page, but Metacritic scores change as new reviews are published, and the figure cited here reflects the score at the time of reporting.
  • Whether the IGF will adopt the anti-AI policy the jurors have urged. The jurors' statement is a request, not a decision, and the festival has not announced a rule change.
  • The full text of the jurors' statement beyond the passages quoted in coverage. The statement was published on Bluesky, and the quotations in this article are taken from reporting of it, not from a direct reading of the original post.

The bottom line

House House used large language models to help fix technical problems in the final stretch of making Big Walk, and it says that use touched well under 0.1% of the game's code and nothing else. It disclosed that use to the IGF when asked, published it on its own website when it became a matter of public interest, and did not disclose it on Steam because Valve's rules do not request disclosure of code or chatbot use. Nineteen IGF jurors have said they will not vote for any game that uses generative AI in any capacity, including code. Players responded with a flood of negative Steam reviews. The studio has apologised, called its lack of an AI policy a failure, and promised to do better.

That is the story, without exaggeration in either direction. It is not a story about a game secretly made by AI. It is not a story about a studio that hid its methods and was caught. It is a story about a small, technical, late-stage use of a chatbot, disclosed through the proper channels, that collided with a principled objection to any AI use at all — and about a studio, a festival and a platform that are all now discovering that the rules they wrote for one world do not fit the world they are in. Whether Big Walk wins awards in March 2027 will be decided by jurors. Whether it is remembered as a great game will be decided by players. And whether the industry learns anything from the controversy will be decided by what studios, festivals and platforms do next.

Similar Posts