Read the claim precisely

Zelnick spoke about generative AI in what Rockstar is building and described the game world as handcrafted. He did not publish an inventory of every internal software tool. Expanding the quote beyond what he said would turn a clear production claim into speculation.

The claim still tells us something important. Rockstar considers authored world-building part of GTA VI's identity and wants that work understood as deliberate. The project also began before the current generation of tools matured, so changing a stable pipeline would have introduced risk into one of the industry's largest productions.

Rockstar's choice does not scale down

Rockstar can support enormous teams, long schedules and an established technical stack. A solo studio or a team of five works under a different economic equation. Using the most heavily resourced studios as proof that everyone should reject labour-saving tools confuses prestige with strategy.

The comparison also runs the other way. AI will not give a small team Rockstar's taste, organisation, capital or distribution. It can shorten implementation, widen exploration and move repetitive work off the critical path. Those gains matter even when they never produce a GTA-sized world.

Draw the boundary at what ships

AI used to search logs, organise production information or generate a disposable test case raises different questions from generated dialogue, characters or locations placed in front of the player. The first changes how the studio works. The second becomes part of the work the studio is asking people to buy.

That boundary is not a moral loophole. Internal tools still need security, legal review and competent oversight. It is a practical way to decide where provenance, disclosure and creative authorship carry the greatest weight. Rockstar's statement sits on the player-facing side of that line.

Faster asset creation does not manufacture taste

Zelnick's stronger argument is that easier assets do not create hit games. He is right. Players already have more games than they can play. Lowering the cost of another image, model or block of code increases supply. It does not decide which idea deserves attention.

That is the part of Zelnick's argument I expect to last. AI can make more candidates cheaper. A studio still has to decide what deserves to exist, reject most of what was produced and carry one coherent decision across years of work. Lower production cost makes that judgement more important because there is more material competing for attention.

What this probably looks like in three years

I expect serious studios to stop discussing AI as one category. Asset generation, code assistance, localisation, production search and live player-facing generation carry different risks. They will end up with different approval paths, disclosure rules and review standards.

Rockstar's hand-built world and Take-Two's hundreds of AI experiments already point in that direction. Studios will protect the work that defines them and automate work where human time adds little. The competitive advantage will come from knowing the difference.

THE WORKING RULE

The important question is no longer whether a company uses AI somewhere. It is where the tool enters the production line, whether its output reaches the player and who remains answerable for the result.

Sources and further reading

Browse more articles