Rise of The Silver Coder
Last week I sat down with a group of gaming VC veterans – people who've backed studios from pre-seed to unicorn – and we ended on the question that sparked a weekend of research and writing: what does AI actually change about the economics of building games?
Not "AI will make games". That's the headline. The real question is underneath it: where does value creation shift when the cost of building compresses but the cost of winning stays the same?
The next day, Anthropic released Claude Opus 4.6. OpenAI dropped their latest model hours later. These aren't incremental upgrades. AI development tools now allow a single developer to orchestrate parallel coding agents against a detailed spec, compressing cycle times and content throughput in ways that weren't possible even six months ago. The timing crystallised something I'd been reflecting on since the conversation: the most expensive layer in game development – the one that absorbs most of the capital, time, and risk – is compressing. And most of the industry isn't structured to benefit.
The GTA VI Paradox
GTA VI keeps getting delayed. GTA V was released in 2013; I've spent half of my life waiting for this moment. Literally.
Let me be precise about why. GTA VI isn't delayed because Rockstar can't use AI. It's delayed because scope expansion, platform transitions, a culture of obsessive polish, and the sheer coordination overhead of 2,000+ people working across a decade-old codebase create compounding complexity that no tool – AI or otherwise – can shortcut. Rockstar's delays are a Rockstar problem, not an industry-wide indictment.
But there's a structural observation underneath the specific case. Rockstar has legacy codebases a decade deep, institutional processes designed for human-scale waterfall development, and coordination costs that compound with every additional team. The transformation layer – the brutal, expensive middle layer between creative vision and shippable product – is calcified by organisational architecture, not by lack of ambition or talent.
Naturally, incumbents are adapting. EA has over 100 active AI projects and a partnership with Stability AI. Supercell is running AI incubators. King replaced roughly 200 staff with AI tools those employees built. These aren't trivial moves. But there's a difference between bolting AI onto existing workflows and building from zero with AI as the foundation. The former optimises; the latter reimagines.
Here's the insight that stuck with me from the conversation: the problem at incumbents isn't talent. It's organisational architecture. The same world-class developers, freed from the coordination tax of a 2,000-person studio, building a lean team that's AI-native from day one – that's a fundamentally different proposition.
And that's not hypothetical. It's happening right now.
What the Transformation Layer Actually Is
Everyone in gaming understands the vision problem is solved. Rockstar knew exactly what GTA VI should be years ago. The problem was always the transformation: 2,000 people, an estimated $2 billion, 8+ years to turn specification into product.
I spoke to senior engineers about where projects actually die. The answer was simple, and it reinforced why I never pursued a CS degree.
The transformation layer is integration code. It's debugging. For the non-coders: this is the "Weapon" vs. "Weapons" variable name mismatch that costs a senior engineer a week to trace through a million-line codebase. It's where projects die, budgets bloat, and timelines slip. Unglamorous, unforgiving, and until now, largely irreducible. Most development spend goes here – not on creative vision, but on the messy, painful work of making that vision run on actual hardware.
Today, AI meaningfully accelerates content creation, prototyping, and boilerplate code generation. It does not yet solve the hardest production problems: systemic integration across complex codebases, platform compliance, live-ops infrastructure, or the iteration-under-uncertainty that defines shipping a real product. Those are coordination and judgment problems, not code generation problems.
But the trajectory matters for investors. What AI is compressing right now – cycle time, content throughput, tooling friction, and the minimum team size required for credible experimentation – is enough to shift the economics of who can build a viable studio. You don't need AI to solve every production problem. You need it to compress enough of the pipeline that a domain expert with taste can get from spec to prototype to telemetry to economy tuning to UA test to iteration – the loop that actually determines whether a game succeeds – without needing a 40-person team to run it.
Your spec is becoming a larger share of your product. And the people best positioned to write that spec aren't 22-year-old CS graduates. They're the 40 and 50-year-old industry veterans who've shipped successful games and know exactly what works.
The Silver Coder
This is the most investable insight I see in the AI-gaming landscape, and the one I think most people are missing.
AI-native tools are compressing the minimum viable team size and shifting the premium from engineering capacity to domain judgment. That's the thesis in one sentence.
Think about the 20-year King veteran who knows exactly which meta-game mechanics drive D30 retention. The Supercell designer who's mapped every live-ops failure mode across three hit titles. The iGaming compliance expert who could design a regulated product across 15 jurisdictions in their sleep. These people always had the domain knowledge. They just couldn't build. The transformation layer stood between what they knew and what they could ship.
Now they can build. AI development tools plus deep domain expertise is the new founding team profile.
The macro data tells the redistribution story. An estimated 45,000 jobs were lost from gaming between 2022 and 2025 (various sources). GDC's 2025 State of the Industry survey reports 21% of developers now working independently, 32% shifting to indie studios instead of AAA, and one in ten losing their job in the past year. The domain knowledge isn't disappearing – it's redistributing. It's leaving organisations that can't capture AI's value and landing in the hands of individuals who can.
The pattern is already producing results. Simon Davis, ex-King and ex-Ubisoft, co-founded Mighty Bear Games in 2017. By mid-2023, the studio reported the vast majority of its content was AI-generated (GamesBeat). His team now ships new Telegram games every two to three weeks, reaching 5.5 million active users through GOAT Gaming. His conviction is absolute: "Small, AI-native teams could outcompete studios hundreds of times their size”.
Pany Haritatos spent two decades in games. Founded a browser studio, sold it to Zynga. Built a mobile studio, sold it to Kongregate. Became CEO of Kongregate, sold the company to MTG. Led Snap's games division. In 2023, he looked at the AI inflection and founded Series AI. Raised $7.9M seed from a16z, then an oversubscribed $28M Series A from Netflix, Dell, and BITKRAFT. In his words: "It was so very clear that AI was about to explode in capabilities. But so many people in the industry thought it was hype." Haritatos is that profile – domain expertise meeting AI capability, with the pattern recognition to time it right.
Or take EasyWin, a Velo Partners portfolio company. A team of eight generating reported annualised GMV of $30 million with strong payer retention. Founder Ivan Leshkevich, ex-Mamboo Games, built a global tournament platform for cash-prize casual puzzle games. It works not because AI is novel, but because Leshkevich has deep domain knowledge in skill-based gaming and AI lets him move without a 50-person team dragging behind.
A fair challenge: Haritatos and Davis are successful operators who adopted AI to accelerate, not domain experts who were previously locked out of building. The purest silver coder – the non-technical veteran who ships their first product using AI tools – is building quietly right now, pre-launch. The signal to watch is specific: over the next 12-18 months, track seed-stage teams where founders have 15+ years of gaming or iGaming operating experience and minimal engineering background. But don't just watch who they are – watch what their early metrics look like relative to burn and time. If these teams are reaching soft-launch with meaningful cohorts on materially lower burn, showing early retention signal (D1/D7/D30 curves comparable to funded peers) and economy tuning competence (stable ARPDAU, payer conversion) – the thesis is confirmed. If domain expertise doesn't translate into faster iteration toward product-market fit, I'm wrong.
That's the silver coder. And for seed-stage VCs, that's the new founder archetype worth underwriting.
The Hard Questions This Thesis Has to Answer
Before the silver coder thesis earns conviction, it must survive some uncomfortable questions. I'd rather ask them myself than wait for someone else to.
If domain experts are advantaged now, why didn't they win when Web2 tools made building easier?
WordPress, Shopify, no-code platforms – they all compressed build costs. The honest answer: they produced a lot of lifestyle businesses and very few venture-scale outcomes. The difference with AI is the magnitude of compression and the complexity of what can now be built. No-code tools let you build a landing page. AI development tools let you build a functional game. The output ceiling is categorically different. But the risk that AI-native studios become lifestyle businesses rather than venture-scale outcomes is real, and any honest investor must sit with that.
If prototyping is cheap, won't capital flood in early and compress returns?
Probably. When the cost of testing an idea drops dramatically, more ideas get tested, more studios get funded, and most will fail. The mobile gold rush post-2012 proved exactly this: cheaper development meant more games, not more winners. The counterargument is that AI compresses not just the build but the iteration cycle – founders with genuine taste and domain knowledge separate from the noise more quickly. But the flood risk is real.
Cheaper creation doesn't mean more successful studios.
This is the criticism I take most seriously. Gaming success is demand-side constrained, not supply-side. Hits are driven by distribution, user acquisition economics, IP, and network effects. Lower development costs historically produce more failure, not more winners. The silver coder thesis is a supply-side argument. It needs a demand-side answer. Mine is this: the silver coder's advantage isn't just cheaper building – it's faster iteration toward product-market fit, combined with domain knowledge that informs what to build, not just how. The King veteran doesn't just prototype faster. They prototype the right thing, because they've spent 20 years learning what players want. Taste plus speed plus domain knowledge is the edge. Cheap production alone is not.
The Cost of Winning Stays the Same
There's a dynamic the supply-side argument can't escape: when building gets cheaper, winning gets harder. Word spreads fast. Discovery gets noisier.
More studios shipping more games means more competition for the same finite attention. Paid user acquisition becomes more important, not less – and CPIs in competitive genres have been trending upward, particularly in mobile and iGaming. When everyone can build, distribution becomes the constraint and cloning accelerates.
If AI compresses your build time, it compresses your competitors too. Novelty windows shrink. The time between launch and first clone shortens. Which means LTV compresses for anyone who can't build network effects, IP moats, or community lock-in.
Platform take-rates persist. Apple and Google still extract 30%. Steam still owns PC discovery. No amount of AI-native development changes the fact that platforms sit between studios and players, extracting margin regardless of how efficiently the studio operates.
This is the demand-side reality the silver coder thesis must coexist with. Building cheaper doesn't make winning cheaper. In fact, it might make winning more expensive, because the marginal studio flooding the market drives up the cost of being found.
Who Captures the Surplus?
This is the question behind all of it, and the one that determines whether the silver coder thesis produces venture returns or just a better class of indie studio: in an AI-native world, who captures the economic surplus – studios, platforms, or tooling?
History says platforms. After the mobile revolution compressed development costs, Apple and Google captured much of the value through distribution control and a 30% revenue share. Studios built more cheaply; platforms extracted more efficiently. If AI replays that dynamic, the silver coder builds a better game faster and still hands 30% to the App Store.
Tooling companies have a claim too. Unity and Unreal captured enormous value as the infrastructure layer of the last generation. If AI-native engines – like Series AI's Rho Engine – become the new standard, the surplus flows to whoever owns the development stack.
But I think there's a case for studios in specific conditions – and this is where the thesis narrows into something investable.
The key variable is whether the studio controls its own distribution and economics. In mobile gaming, the App Store owns distribution and UA economics dictate margins. Studios are price-takers. In iGaming, the dynamics are structurally different – but the nuance matters.
In iGaming, "studio" often really means operator. The silver coder who is most advantaged isn't just building a game – they're building or operating within a vertically integrated stack: player acquisition, compliance, payment infrastructure, CRM, and retention. An operator with domain expertise in licensing, player lifecycle management, and responsible gambling can build direct player relationships, own their data, and control their economics in ways a mobile studio on the App Store simply cannot.
But a pure content supplier in iGaming – building games for operators to distribute – faces many of the same extraction risks as a mobile studio on a platform. Affiliates, payment processing costs, and operator revenue shares all eat margin. The difference is that these dynamics are more segmented than in mobile: affiliate dependency varies if you build brand and direct acquisition channels, payment costs vary enormously by geography and method, and operators who invest in CRM and retention can shift margin structure meaningfully.
This is why the silver coder thesis is sharpest in regulated markets – specifically for operator-founders, not just content suppliers. Not just because domain expertise is more defensible, but because the value chain allows the right kind of studio to keep more of what it creates. Regulatory moats protect against both the content flood and platform extraction. The silver coder in iGaming isn't just a better founder archetype. They're operating in one of the few market structures where AI-native economics and defensible margins can coexist.
The Counter-Thesis Deserves Respect
Content flood is real. Steam saw AI-flagged titles surge – by some estimates reaching 8,000 in the first half of 2025, up from roughly 1,000 in all of 2024, representing around 20% of new releases (Totally Human Media, VGC). Platform gatekeepers are tightening: Steam now requires AI content disclosure, Apple is restricting how apps share data with third-party AI. Community backlash against "AI slop" is growing. And the creative sceptics make a fair point – Konrad Tomaszkiewicz, director of The Witcher 3, says flatly: "Games created with only AI will not have soul”.
He's right. The gaming VC veterans were right. And that's exactly the point.
AI raises the floor. Anyone can build. But it simultaneously raises the ceiling – the best operators, the ones with genuine taste and domain expertise, move faster than ever. The cost of testing whether you have taste just dropped dramatically. You can now ship a prototype, find out if your instinct was right, and iterate – all before a traditional studio finishes its pre-production sprint.
The gap between floor and ceiling is where VCs make money.
The New Diligence Question
The transformation layer is compressing. That's the structural fact. What follows is a shift in how gaming VCs should evaluate seed-stage opportunities.
The old question was "can this team build?" The new question is "does this team know?". When AI compresses the distance between specification and product, domain expertise becomes the scarce input. Technical execution is table stakes. Taste, conviction about what players want, understanding of distribution and UA economics, knowledge of regulatory environments, instinct for live-ops failure modes – these are the moats now.
But "does this team know?" is only half the diligence. The other half is “does this team operate in a market structure where they can capture the value they create?”. In open-platform gaming, that's hard. In regulated markets with direct distribution and ownable player relationships, it's possible.
The best version of the silver coder thesis isn't just a founder profile. It's a founder profile matched to a market structure.
Incumbents have the knowledge but can't restructure. Their best people know it, and they're leaving. Insurgents with domain expertise start AI-native, in markets where regulation protects margins and distribution is ownable.
That's the bet.