PH3 CTO on AI in Game Development: Big Productivity Gains, but Human Review Still Needed

PH3's CTO argues that frontier LLM agent systems in 2026 are genuine productivity boosters, but relying on them without review or experience can produce sub-optimal outcomes even in isolated tasks. He cites a coding agent, ranked #1 to #4 on coding benchmarks, that implemented an achievement check by iterating over up to ~11000 game entities every simulation frame when a far cheaper approach existed. He also notes games are currently less directly affected by AI than other software fields because fast feedback loops and clear objective functions are harder to establish.