Take-Two's AI Mistake Wasn't Hiring the Wrong People. It Was Firing the Right Ones.

SOURCE THREAD: x.com/davidsenra/status/2056011858499318035

The analyzing-text breakdown of the conversation

Unity

Take-Two's choice to slot AI entirely under the CTO, frame it as infrastructure, and dissolve the dedicated AI team six weeks earlier is rational at the governance layer and corrosive at the leverage layer, because it confuses tool deployment with creative mastery and removes the very people who could have bridged the two.

Terms (specialized usage)

• Infrastructure. Zelnick's reduction of AI to "data sets, compute, and a large language model mushed together." Used to argue AI belongs under a CTO the same way databases or render pipelines do. The conversation contests this definition.
• Deployment. Issuing enterprise ChatGPT and Claude licenses across the workforce. Treated by the CEO as substantially equivalent to having an AI capability.
• Mastery. The non-obvious, weird, frontline experimentation that turns a commoditized tool into a competitive edge. Argued in this conversation to be a different and scarcer thing than deployment.
• Hit creation vs asset creation. Zelnick's own pair. Asset creation is producing textures, code, and other inputs; hit creation is producing something culturally surprising. He concedes AI is good at the former, useless at the latter.
• Forward-looking context. What domain creatives supply that backward-looking data sets cannot. The bridge that turns raw model output into something non-derivative.
• Pandora's box. Kent's metaphor for AI as a capability that, once opened, cannot really be controlled, only contained. Slotting AI under the CTO and dissolving the AI team reads as an attempt to close the lid rather than work with what was let out.
• Board heel. The public-company gravitational pull that rewards centralization, predictability, and clean accounting over open-ended experimentation. Frames the org choice as politics, not strategy.
• It hits different. Kent's blunt insistence that AI is unlike past technology categories because it has taste, opinions, and creative output baked in. Whoever owns it inside a company is shaping the company's creative output, whether they know it or not.
• Bridge function. The role played by technical specialists who are simultaneously artists and engineers, embedded inside the creative pipeline rather than above it. Catmull's RenderMan group at Pixar; Weta's pipeline TDs on Avatar; Take-Two's now-dissolved AI team. The function exists to translate raw technical capability into something a creative director can actually deploy at the shot or scene level.
• Creative R&D shop vs corporate IT. Catmull's distinction, lifted into this conversation. Same headcount, same budget line, completely different operating model. One invents inside the work. The other governs from the side.
• Bifurcation. Splitting AI ownership into two distinct roles instead of one: infrastructure, governance, security, and procurement go to the CTO; creative experimentation, including unauthorized experimentation, lives embedded inside the studios. The conversation argues this was the actually-available better move.
• Spin-out as protection. Letting the AI team continue as a separately-stood-up R&D entity, the way Pixar eventually spun RenderMan out as its own business. Frames AI research on "how games are made" as a business question, not a tech question.
• Tool plus hand. The proposition that creativity does not live in the tool, whether pencil, brush, or LLM, but in the hand using it. Reframes what counts as a creative, and exposes the unstated assumption inside Zelnick's choice to fire the AI team.

Propositions (claims made by the conversation)

1. Take-Two rolled out enterprise ChatGPT and Claude across the company.
2. Take-Two built a dedicated AI team led by Luke Dicken between 2024 and early 2025.
3. In April 2026 Take-Two laid off Dicken and a significant portion of the AI team.
4. Zelnick has now formally placed all remaining AI work under the new CTO, Dave Klein.
5. Zelnick frames AI as technology infrastructure on par with databases or render pipelines.
6. Zelnick believes AI is good at asset creation but cannot produce cultural hits.
7. Zelnick also says he hires elite creatives and gets out of their way.
8. Deployment of an AI tool is not the same as mastery of its creative use.
9. IT and CTO functions have a documented historical pattern of stifling creative experimentation with new tools.
10. The dissolved AI team was the bridge between model capability and Rockstar / 2K creative workflows.
11. Public-company governance pressure rewards centralization, predictability, and de-risked execution.
12. The org choice (centralize + dissolve AI team) directly contradicts Zelnick's stated "hire creatives and get out of their way" philosophy.
13. AI is not inert in the way databases are inert: it has taste embedded from training data and exports that taste into every output.
14. Roughly 12% of enterprises now run fully centralized AI; the majority moved to hybrid or embedded ownership.
15. The competitive moat in the AI era is not which models you license but who inside your company is allowed to play with them first.
16. Pixar's Toy Story (1995) was produced by a team that built its own tools, including RenderMan, because nothing off the shelf could do what the directors wanted.
17. Pixar's technical directors functioned as artists and engineers simultaneously, embedded shot by shot inside the production pipeline.
18. Ed Catmull ran Pixar's technology as a creative R&D shop, not as corporate IT.
19. Avatar (2009) followed the same pattern at larger scale: Cameron and Weta Digital invented or heavily customized the Simulcam virtual camera, the facial-capture rig, and deep-compositing workflows.
20. The Avatar breakthroughs came from VFX supervisors, technical directors, and pipeline TDs working shoulder to shoulder with the director and cinematographers, not from a centralized IT function.
21. Across both Pixar and Avatar, IT and procurement standardized and scaled the toolchain only after creatives had proven what worked.
22. The Take-Two AI team that was dissolved in April 2026 was performing exactly the bridge function that produced Toy Story and Avatar.
23. The available alternatives to Take-Two's actual move were (a) decentralize the AI team and bifurcate the function and (b) spin the AI team out as a separate R&D entity.
24. Bifurcation would have placed infrastructure, governance, security, and procurement under the CTO while pushing creative experimentation, including unauthorized experimentation, into the studios.
25. Spin-out would have treated AI research on "how games are made" as a business question requiring wild creativity, not a tech question requiring containment.
26. Either alternative would have protected the creatives Zelnick says he wants to protect.
27. The chosen move is only internally consistent if Zelnick does not see AI-fluent specialists as creatives.
28. A tool, whether a pencil or an LLM, is inert without a hand on it; creativity lives in the hand, not in the tool.

Arguments

Argument A: Deployment is not mastery.

• Premise 1: Every competitor can license the same enterprise Claude and ChatGPT seats in the same week.
• Premise 2: Competitive advantage requires doing something competitors cannot easily replicate.
• Premise 3 (omitted but doing the work): Historically, the non-replicable use of a new tool emerges from frontline creatives experimenting in unauthorized ways before IT formalizes anything (desktop publishing, the web, Figma, Notion). Inside the entertainment industry specifically, the leverage came from embedded technical-creative hybrids: Pixar's RenderMan team on Toy Story (1995), Weta's pipeline TDs on Avatar (2009), both run as creative R&D shops rather than corporate IT.
• Conclusion: Issuing licenses produces no advantage. Mastery is the scarce variable, and it does not emerge from CTO-led standardization.

Argument B: Backward-looking data needs forward-looking taste (Zelnick's own).

• Premise 1: AI is built on data sets, which by their nature look backward.
• Premise 2: Cultural hits require forward-looking instinct that data cannot supply.
• Conclusion: AI cannot produce hits on its own; humans with taste still must drive it.
• Kent's implicit counter (omitted step Zelnick does not name): The team Take-Two laid off was specifically the bridge between human taste and AI tooling. If forward-looking taste is the scarce input, dissolving the team that married taste to tooling is the move you make last, not first.

Argument C: AI is not just technology.

• Premise 1: Traditional infrastructure (databases, render pipelines) is inert until pointed at a problem.
• Premise 2: AI carries embedded taste, opinions, and stylistic preferences from training.
• Premise 3: Whoever decides how AI is used inside a company is shaping the company's creative output, whether they realize it or not.
• Conclusion: Slotting AI under a function whose mandate is standardization, security, and cost control means a security mindset is now silently shaping creative output. That is a category error, not a governance choice.

Argument E: The better move was available.

• Premise 1: Three options were on the table: centralize and fire, decentralize and bifurcate, or spin out.
• Premise 2: Decentralization splits AI ownership into infrastructure (CTO) and creative experimentation (embedded inside the studios), preserving both governance and creative leverage.
• Premise 3: Spin-out treats AI research on "how games are made" as a business in its own right, the way Pixar treated RenderMan.
• Premise 4: Both alternatives would have protected the creatives Zelnick says he wants to protect.
• Conclusion: The chosen move is not the only feasible option. It is the option a CEO picks when he does not see AI-fluent specialists as creatives. The constraint is the mindset on what creativity is, not the org chart.

Argument D: The Pandora's box read (the political-economy argument).

• Premise 1: Public-company boards reward predictability and penalize visible volatility.
• Premise 2: An autonomous AI team running open-ended experiments looks volatile.
• Premise 3: Reframing AI as infrastructure under a CTO is the cleanest political move to declare the topic "handled."
• Premise 4 (omitted step Zelnick does not say out loud): The April layoffs and the May org-structure announcement are the same act in two scenes.
• Conclusion: This is risk management dressed up as strategy. Rational at the governance layer. Corrosive at the leverage layer.

Claims (factual)

• Take-Two has "hundreds" of AI projects underway.
• Dave Klein is the new CTO; all AI reports to him.
• Luke Dicken (Head of AI) and an undisclosed but significant portion of the AI team were let go in April 2026.
• GTA 6 launches November 19, 2026.
• GTA 6 was built without generative AI in its core creative development.
• Mature enterprises have largely moved away from fully centralized AI ownership.
• Pixar wrote RenderMan in-house for Toy Story (1995); the technical directors operated as embedded artist-engineers.
• Weta Digital and James Cameron invented or heavily customized Simulcam, the facial-capture rig, and deep-compositing workflows for Avatar (2009).
• In both Pixar and Avatar, IT and procurement standardized the toolchain only after the creatives had proven what worked.

Questions surfaced

• Is AI really just technology?
• Is the CEO closing the lid on his own Pandora's box?
• Who inside Take-Two now bridges human taste and AI tooling?
• What happens to the next generation of hits after GTA 6, if the bridge team is gone and the remaining AI sits under standardization-minded ownership?
• What should a private company or startup learn from this, and what should it explicitly refuse to copy?
• What does it say about Zelnick's working definition of creativity that he did not see his AI team as creatives worth protecting?

The X article

When a public-company CEO calls AI "just technology" and hands the whole thing to the CTO, that is not strategy. That is risk management wearing a slide deck.

Take-Two did exactly this. Strauss Zelnick told David Senra that AI sits "within the technology world" under their new CTO, because that is what it is to him: data sets, compute, and a large language model mashed together. Enterprise ChatGPT for everyone. Enterprise Claude for everyone. Hundreds of pilots. Lid closed.

Six weeks earlier, they quietly dissolved the dedicated AI team. The head of AI was gone. So were the people who had spent a year wiring machine learning into actual Rockstar and 2K game-development workflows. Stated reason: shifting priorities from upper management.

Read those two moves in sequence and the picture is hard to miss.

Deployment is not mastery.

Installing Claude across your workforce is the easy part. It is also the part that produces no competitive advantage, because every competitor can buy the same licenses in the same week. The hard part, the part that actually wins, is figuring out how designers, writers, artists, and operators bend the tool into shapes nobody else has tried. That work does not happen inside a CTO function whose mandate is standardization, security, and cost control. It never has.

Look at where the actual hits in this industry came from.

Pixar's Toy Story in 1995 came out of a team that built its own tools because nothing off the shelf could do what John Lasseter, Ed Catmull, and the technical directors wanted. RenderMan was written in-house. The lighting, shading, and rendering choices were made shot by shot by people who were simultaneously artists and engineers, embedded inside the production pipeline. Catmull ran the technology, but he ran it as a creative R&D shop, not as corporate IT.

Avatar in 2009 is the same pattern at bigger scale. Cameron and Weta Digital invented or heavily customized the tools they needed: the Simulcam virtual camera, the facial-capture rig, the deep-compositing workflows. The breakthroughs came from VFX supervisors, technical directors, and pipeline TDs sitting shoulder to shoulder with the director and cinematographers. Not from a centralized IT department telling everyone which render farm policy to follow.

The leverage came from technical specialists embedded inside the creative work, given room to invent. IT and procurement showed up later to standardize, scale, and secure what the creatives had already proven worked.

That is exactly the bridge function Take-Two just dismantled.

And the same pattern shows up outside film. Desktop publishing did not get its leverage from IT. Neither did the web. Neither did Figma. Neither did Notion. Every meaningful creative-tool wave was driven by frontline people running unauthorized experiments until the magic showed up, and only then did IT come in to put fences around it.

Take-Two just inverted the order. They built the fences first. Then they fired the only people on payroll whose explicit job was to find the magic.

AI is not just technology. It hits different.

A database is inert. A render pipeline is inert. They do nothing until someone points them at a problem. AI is not inert. It carries taste, opinions, and stylistic preferences from its training data, and it exports those preferences into every output your team produces, quietly, all day, forever. Whoever decides how AI is used inside your company is shaping the company's creative output, whether they realize it or not.

If that person is your CTO, your creative output now reflects what a security-and-uptime mindset thinks AI should produce. Which is asset creation, efficiency gains, and derivative work. Exactly the things Zelnick himself said do not produce hits.

His own argument lands on his own decision. He just did not name it.

This is what board heel looks like.

Public-company CEOs cannot escape board gravity. Boards want predictability. AI experimentation looks volatile. The cleanest political move is to declare AI "infrastructure," put it under the CTO, point at the enterprise licenses on the procurement bill, and tell investors the topic is handled. Risk closed. Box shut.

That move is rational at the governance layer. It is a strategic mistake at the leverage layer. The two are not the same and confusing them is the failure pattern.

The better move was sitting right there.

The choice was not "centralize or fire." Two better moves were on the table. He took neither.

Move one. Decentralize the AI team. Keep them on payroll. Split the function in two. Infrastructure, governance, security, and procurement go to the CTO, where they belong. Creative experimentation, including the unauthorized kind, goes to the studios, embedded inside the Rockstar and 2K production pipelines. Two separate jobs. Two separate reporting lines. One sees AI as plumbing. The other sees it as paint. Both true. Both needed.

Move two. Spin the AI team out. Let them keep researching how games are made. That is not a tech question. That is a business question, and it requires wild creativity. Pixar already wrote the playbook for this. Catmull's group invented RenderMan in service of one movie, then carried the toolchain forward as a business in its own right. Take-Two had the same option with the bridge team they just fired.

Either move would have actually protected the creatives Zelnick says he wants to protect. Both would have required him to change his mindset on what creativity is.

A pencil in the right hands creates wonder at scale. So does an LLM. The tool is not the thing. The hand on the tool is the thing. And a hand on an LLM is just as much a creative as a hand on a pencil, if you have the imagination to see it.

He does not.

What private companies and startups should not copy.

If you are a founder reading this and your current AI plan is "issue Claude logins, hand the rest to IT, declare victory," stop.

Decentralize. Embed AI specialists inside creative and product teams, not above them. Let frontline operators run weird experiments before you build the governance fences. The governance comes later, after you know which experiments are worth fencing. Reverse that order and you will end up with a tidy policy document and zero competitive advantage.

Take-Two made the safe public-company move. You are not Take-Two. Do not make it.

What would change at your company if you put AI in the hands of your weirdest creatives instead of your tidiest CTO?

Shorter cuts

Short tease tweet

Take-Two just handed all of AI to the CTO "because it's technology."

Six weeks earlier they laid off the team that was building AI into actual game-dev workflows.

That is not strategy. That is risk management dressed up as one.

Deployment is not mastery. AI is not just technology. It hits different.

Mid-length variant

Take-Two's CEO just told David Senra that AI sits under the new CTO "because it's technology." Enterprise ChatGPT for everyone. Enterprise Claude for everyone. Hundreds of pilots. Lid closed.

Six weeks earlier they dissolved the dedicated AI team. The head of AI was gone. So were the people who had spent a year wiring ML into Rockstar and 2K workflows.

Read the two moves in sequence.

Deployment is not mastery. Issuing licenses is the easy part. Every competitor can do the same thing in the same week. The hard part, the winning part, is frontline creatives bending the tool into shapes nobody else has tried. That work has never come from a CTO function whose job is standardization, security, and cost control.

Look at where the hits in this industry actually came from. Toy Story shipped because Pixar wrote RenderMan in-house and let technical directors make shot-by-shot choices alongside the artists. Avatar shipped because Cameron and Weta invented Simulcam and the facial-capture rig with pipeline TDs sitting next to the director. Catmull ran the technology as a creative R&D shop, not as corporate IT. That is the bridge function Take-Two just dismantled.

Same pattern outside film. Desktop publishing did not get its leverage from IT. Neither did the web, Figma, or Notion. Every creative-tool wave was driven by people on the ground running unauthorized experiments first, then IT came in to fence the magic.

Take-Two inverted that. They built the fences, then fired the people whose job was to find the magic.

AI is also not inert. It carries taste from its training data and exports that taste into every output your team makes. Whoever owns it inside your company is shaping your creative output, knowingly or not. Hand it to a security-and-uptime mindset and you get derivative work. Which is exactly what Zelnick himself said does not produce hits.

This is rational at the governance layer. It is a strategic mistake at the leverage layer.

Two better moves were on the table. Decentralize the team and bifurcate the function: infrastructure to the CTO, creative experimentation embedded in the studios. Or spin the team out and let them keep researching how games are made. Either move would have protected the creatives Zelnick says he wants to protect. Both required him to change his mindset on what creativity is.

A pencil in the right hands creates wonder at scale. So does an LLM. The tool is not the thing. The hand on the tool is the thing.

Founders, do not copy the public-company move. You are not Take-Two.