Opinion: Why AI NPCs with Claude 3.5 Are a Gimmick and Not Worth the Cost for Indie Games

The games industry is in the middle of an AI hype cycle, and indie developers are increasingly being pressured to adopt large language model (LLM) powered NPCs, with Anthropic’s Claude 3.5 Sonnet often cited as the top choice for natural, context-aware dialogue. But for small, self-funded or crowdfunded indie teams, this trend is a dangerous distraction: Claude 3.5 AI NPCs are an overpriced gimmick that deliver negligible player value while draining already stretched budgets.

The Hidden Cost of LLM NPCs

First, let’s talk numbers. Claude 3.5 Sonnet’s API pricing is $3 per million input tokens and $15 per million output tokens as of late 2024. For a single NPC interaction averaging 500 input tokens (player query + context) and 200 output tokens (NPC response), that’s roughly $0.0045 per interaction. That sounds negligible until you scale it: a small indie game with 5 unique AI NPCs, 2,000 monthly active players, and an average of 8 NPC interactions per player would rack up $360 in monthly API costs, with no cap on spend if the game gains traction. Traditional scripted NPCs, by contrast, require a one-time investment of 10–20 hours of writing and implementation per NPC, costing a fixed $500–$1,000 total for 5 NPCs even at modest indie contractor rates. The LLM option becomes a recurring, uncapped expense that grows with your player base, while scripted NPCs are a fixed, predictable cost.

Technical Friction Indie Teams Can’t Afford

Beyond cost, Claude 3.5 introduces technical hurdles that are uniquely punishing for small teams. LLMs require persistent internet connectivity to function, meaning any indie game using them can’t support offline play, a feature that’s table stakes for many indie titles targeting Steam Deck, console, or regions with spotty internet. Latency is another issue: even with optimized API calls, players will face 1–3 second delays for NPC responses, breaking immersion in fast-paced or narrative-heavy games. Then there’s the context window problem: Claude 3.5’s 200k token context window sounds large, but for NPCs that need to remember player choices across a 10-hour game, you’ll either have to truncate context (leading to NPCs forgetting key plot points) or pay for even more tokens to store persistent state. And unlike scripted NPCs, which behave exactly as written, LLM NPCs hallucinate: they might break lore, reveal spoilers, or say offensive content that indie teams don’t have the QA bandwidth to catch at scale.

Players Don’t Want This (And Indie Games Don’t Need It)

The biggest myth driving this trend is that players crave dynamic, unscripted NPC interactions. In reality, indie game success is built on curated, intentional design: players remember a well-written, consistent side quest NPC more than a generic AI character that rambles about the weather. For narrative-driven indies, scripted NPCs ensure that every line of dialogue serves the story, worldbuilding, and gameplay goals. LLM NPCs, by nature, are unpredictable, which undermines the tight, cohesive experiences that indie games are known for. A 2024 survey of indie game players found that only 12% of respondents would pay more for a game with AI NPCs, while 68% said they’d avoid a game that required always-on internet for core features.

Opportunity Cost: Where That Money Should Go

Indie budgets are almost always constrained: a typical small indie game has a total budget of $50k–$200k. Spending $5k+ per year on LLM API costs (a conservative estimate for a modestly successful game) is money that could go to 2–3 months of additional art production, a full sound design pass, or a targeted marketing campaign that actually drives wishlists. For solo devs or tiny teams, that tradeoff is even more stark: every hour spent integrating, debugging, and monitoring LLM NPCs is an hour not spent on core gameplay systems, bug fixes, or polish.

Conclusion

Claude 3.5 is an impressive tool for enterprise applications, content creation, and prototyping. But for indie games, it’s a solution in search of a problem. AI NPCs built on LLMs are a gimmick that offer no meaningful improvement to player experience while introducing uncapped costs, technical debt, and design risks that small teams can’t absorb. Indie devs should ignore the hype, stick to proven, scripted NPC systems, and focus their limited resources on the things that actually make games great: tight gameplay, compelling stories, and polished production value.