In 2008 Neil Druckmann wrote up Naughty Dog’s first PlayStation 3 project — internally codenamed “Big” — for Game Developer magazine. It is a textbook five-right / five-wrong postmortem, and what makes it worth reading today is that almost every failure is a tools or pipeline failure. Nothing here is about art or story; it is about a studio that got too clever with its own infrastructure.

What went right:

  • They planned up front, then ditched the plan when prototyping said so — gunplay was specced as auto lock-on, months went into lock-on targeting, and none of it gelled until they switched to free manual aiming. They made the call “just early enough to be practical.”
  • Gameplay was deliberately narrow compared to Jak and Daxter, and built systemically — so refining the player move set late in production improved nearly every level at once.
  • Same actors performed both mocap and voice, with table reads, blocking walk-throughs, and improv on set. Characters were designed plain on purpose (“Drake’s outfit is really no more than a t-shirt and jeans”) so personality came from performance rather than accessories.
  • Animation and AI were initially scoped as general solutions, then cut down. The shipped system layered and blended full-body, partial, and additive animations — sometimes up to 25 layered Drake animations at once — plus inverse kinematics to register him to the environment.
  • Several purpose-built renderers ran at once (geometry, water, particles, decals, HUD), with water color computed from optical and physical principles rather than textures.

What went wrong:

  • Early months went to engine, rendering, and tools, leaving no programming support for gameplay prototypes. Morale took the hit until the design was focused and communicated.
  • New tools were over-engineered. The first E3 teaser trailer exposed them, and a build pipeline that let people actually run and play levels arrived very late — forcing a retreat to their Jak and Daxter methods.
  • A planned interchangeable body/clothing parts system for enemies blocked the entire enemy implementation until they abandoned it.
  • Realism fought the design: enemies that soaked many hits felt wrong, forced constant retuning and extra spawn waves, and subtle enemy-class silhouettes were unreadable at camera distance.
  • Too few level-layout iterations before final art. Their finding: polish the level while it is still visually simple and you redo less environment work later.

The through-line is stated outright: “getting on with making the game is the best way to make it… do enough planning, but don’t over-plan.” Every ambitious abstraction in this postmortem cost them months, and every recovery was a retreat to the simpler version that unblocked the team.

The level-layout point is the most transferable for small teams. Iterating on untextured geometry is cheap and is the only window where designers, concept artists, and background artists can reconcile ideas; once finished art lands, every change is expensive.