Case Study: BioShock Infinite's Elizabeth — Designing a Companion Who Helps

The design meeting that preceded BioShock Infinite (2013) is a meeting you can imagine if you've ever worked on a companion character. Irrational Games had committed, publicly, to a premise that scared the AI team: the player would spend most of the game accompanied by a young woman named Elizabeth, whom they would grow to care about, and who would be present in nearly every combat encounter.

Every AI programmer in the meeting would have known the history. Companion AI had been the graveyard of otherwise good games for two decades. Ashley in Resident Evil 4 had become an internet meme for frustration. Navi in Ocarina of Time — "Hey, listen!" — had become a meme for the opposite kind of failure, the ally who annoys without helping. Ico's Yorda had been widely praised as a companion, but she was also helpless and slow enough that the entire game was built around protecting her, which is the opposite of what BioShock Infinite wanted. Every attempt at "competent combat partner" in a first-person shooter — Alyx in Half-Life 2, Natalya in GoldenEye, the various friends in Halo — had required heroic amounts of engineering and still produced bugs in every shipping build.

Creative director Ken Levine and his team knew they could not make Elizabeth a conventional companion. They decided, instead, to redefine what a companion was. She would not fight alongside the player. She would not need to be protected. She would not follow rules of physics that would block or kill her. She would be, in design terms, a friendly user interface wearing the skin of a character.

This case study walks through the specific design decisions that made Elizabeth work, and through the tradeoffs those decisions imposed.

The Core Pillars

Irrational's team — principally lead designer Bill Gardner, writer Drew Holmes, and a dedicated AI/animation team led by Chris Kline — committed to five principles for Elizabeth. They are worth listing cleanly because they capture what companion AI should aspire to:

  1. She must never block the player. If Elizabeth is in your way, you should not feel her. She either moves out of the way or phases through you.
  2. She cannot die. No enemy targets her. She takes no damage. Her HP is infinite, and the player never has to look at her health bar.
  3. She provides resources at the moment of need. When the player is low on ammo, she tosses ammo. When the player is low on health, she tosses a medkit. When the player is low on salts (magic), she tosses salts. The timing is calibrated to need, not to interval.
  4. She feels present. Her animation, voice lines, and gaze tracking communicate that she is paying attention to what the player is paying attention to. She looks where the player looks. She reacts to events the player reacts to.
  5. She has narrative agency through gameplay. She opens tears — portals to parallel realities — that change the combat environment. The player asks her to do it by pointing at a tear in the world.

Notice what these principles collectively give up: Elizabeth is not a tactical agent. She does not kill enemies. She does not make independent decisions that change the battlefield on her own initiative. The designers traded autonomy for reliability. In exchange, Elizabeth is reliably present and reliably helpful, which is what the game needs her to be.

Invisible Decisions

The first time you play BioShock Infinite, you do not notice that Elizabeth's collision phases. You do not notice that her resource tosses are timed to your need. You do not notice that she is teleporting past you when you turn around. These decisions are deliberately invisible. Their invisibility is the craft.

Collision

Elizabeth has no collision with the player. This is the first and most foundational decision. In a corridor, when you stop, she does not bump into you; she slows and moves around. When you turn around quickly, she does not block your path; she's already been teleported a step behind.

The teleportation is the second trick. When the player is not looking at Elizabeth — when she is off-screen or outside the player's field of view — the AI can reposition her invisibly. The rule is "never teleport where the player can see you." If the player pivots toward Elizabeth, the last known position stays fixed (because otherwise she would appear to jump). If the player pivots away, Elizabeth repositions to wherever she needs to be next.

In a typical level, Elizabeth is teleported dozens of times per minute. Most of these teleports are invisible. A few are visible, and the team tuned these carefully — a teleport that happens while the camera is mid-pan is less noticeable than one that happens at rest.

Pathfinding

Elizabeth still uses pathfinding (navmesh-based, like most 2013 AAA characters), but the pathfinding is aided by the teleport system. When a standard path would require her to cross the player's path in an awkward way, the system instead teleports her around. When she would normally fall behind, she catches up by teleporting out of sight.

This is the key insight: pathfinding is hard, and navmesh AI will always look occasionally weird (stuck on a corner, orbiting an obstacle, oscillating at a doorway). Elizabeth's system accepts this and layers teleports on top. The navmesh handles 80% of cases; the teleport handles the other 20%.

Resource Tossing

The ammo-and-health toss system is the clearest example of designing against need rather than interval. Here is roughly how it works (reconstructed from designer talks and shipping-game observation — not leaked source):

  • The game tracks the player's resource state continuously: current ammo per weapon, current health, current salts.
  • Resource levels cross thresholds: 30% health, 15% health, 10% ammo, etc.
  • When the player crosses a threshold downward during combat, a counter starts. If the resource stays low for N seconds and combat continues, Elizabeth is flagged to toss.
  • Elizabeth checks line of sight to the player. If she can see the player, she throws; if not, the teleport system repositions her first.
  • The throw has a specific arc and animation. It always lands at the player's feet or in a position they can reach in a second or two.

The effect, from the player's perspective: you run out of ammo, you think "oh no," and then Elizabeth yells "Here!" and tosses you exactly what you need. The feeling is uncanny. You feel seen. A human companion would do this — a good human companion would read your situation and help.

Elizabeth's delivery is not random. It is not interval-based. It is need-driven. That is what makes it feel alive.

The Look-Tracking System

Elizabeth's gaze follows the player's gaze. If the player looks at a poster on the wall, Elizabeth's head turns slightly toward it. If the player looks at an enemy, Elizabeth's head turns toward the enemy. This is handled by a shared attention mechanism: both the player's aim vector and Elizabeth's gaze target read from the same "point of interest" data, with Elizabeth's gaze lagging slightly behind.

The effect: when you and Elizabeth are in a room together, you feel like you are together in the room. She is not a companion on rails, staring straight ahead. She is someone who is looking at the world with you.

This pipeline was a significant animation and engineering investment. The gaze had to blend smoothly across animations. It had to handle cases where Elizabeth was animated-locked (during scripted moments). It had to avoid uncanny-valley failures (staring too long, snapping too fast). The team iterated on it through production.

Voice Barks

Elizabeth has hundreds of barks, triggered by situation: combat events ("You got him!"), exploration events ("Is that a Vox tear?"), damage events ("Are you okay?"), resource events ("Here, catch!"). The barks are what give her a personality beyond the animation.

Irrational's VO budget was significant. Courtnee Draper, who voiced Elizabeth, recorded thousands of lines. The lines were layered — Elizabeth says one thing in this specific encounter, another in the same encounter on replay, a third in a variant — so that the player rarely hears the same line twice in close proximity.

The Design Tradeoffs

Every design decision pays a cost. Elizabeth's design pays specific ones.

She cannot surprise you tactically. Elizabeth cannot flank an enemy, cannot outplay the player, cannot take initiative in combat. Her behavior is reactive and scripted. The player cannot be genuinely impressed by her combat intelligence because she does not have any.

She is not a second character in the game world. Elizabeth exists in a different physical register than the enemies. She phases through physics; she teleports; she does not bleed. A player who pays close attention will notice that Elizabeth operates by different rules, and this can be jarring if the game's narrative asks you to treat her as a peer.

The resource system creates a subtle crutch. Because Elizabeth tosses resources when the player is low, the player never quite experiences running out. Some players felt this blunted the tension of combat in Infinite compared to earlier BioShock games, where resource scarcity was a pillar of the design. The tradeoff was intentional: Irrational wanted the narrative companion to produce a specific cared for feeling, and they were willing to soften combat's resource pressure to get it.

It is expensive. Elizabeth's implementation — the animation pipelines, the teleport system, the barks, the gaze tracking — was a major engineering investment that crowded out other features. Most studios don't have the budget to build a companion to this standard. This is partly why Elizabeth is rare; the pattern is expensive to replicate.

She is not an agent; she is a device. Some critics found Elizabeth less compelling because she has no inner life expressed through gameplay — her choices are scripted, her reactions predetermined. Her character lives in the cutscenes and barks, not in her moment-to-moment play. If you care about "companion characters as agents," Elizabeth is less interesting than The Last of Us's Ellie, who has more simulated independence.

What This Means For Your Design

The Elizabeth approach works when:

  • The companion's narrative role matters more than their tactical role.
  • The player should feel cared for rather than accompanied.
  • The studio has the budget for extensive animation, VO, and engineering.
  • The companion is present across most of the game (not a cameo).

The Elizabeth approach does not work when:

  • The player needs to feel tactical pressure from an ally ("we're in this together").
  • The game is mostly stealth and you want the ally to have real agency (hiding, being spotted).
  • The budget is indie-level.
  • The companion should vary between present and absent meaningfully, with real consequences.

For your own game: if you want a companion and you can afford it, build toward the Elizabeth pattern. Start with "cannot block, cannot die, cannot miss," even if you end up loosening some of these constraints. It is much easier to loosen constraints than to add them back.

Contrast: Ashley in Resident Evil 4 (2005)

The original Ashley is the anti-Elizabeth. Her design premise is "escort NPC — protect her while fighting." She has collision, can be grabbed, takes damage, and has to be protected from enemies throughout significant chunks of the game.

The difference in design philosophy is total:

  • Ashley is a liability to the player; Elizabeth is a resource to the player.
  • Ashley requires the player's attention and effort; Elizabeth provides for the player's needs.
  • Ashley's AI can visibly fail; Elizabeth's design prevents most failure modes from being visible.
  • Ashley's mechanics produce tension through risk; Elizabeth's mechanics produce relief through provision.

The Resident Evil 4 Remake (2023) walked back much of Ashley's original design. She can no longer be grabbed easily. Her pathing was polished. Her AI timing was improved. Capcom did not go full-Elizabeth — Ashley still has collision and can still be hurt — but they moved significantly in that direction, cutting the friction that had made the 2005 version a meme. The remake's reception suggests the adjustment was correct.

Takeaways

  • Elizabeth's design treats the companion as a friendly UI wearing a character skin, not as an autonomous agent in the world.
  • The core decisions — no collision, no death, no missed resources, looking where the player looks — are inexpensive individually but their combined effect is transformative.
  • Invisible teleportation and event-timed resource tosses are the two most important technical tricks.
  • The design pays real costs in tactical authenticity and implementation expense, but produces a feeling of being cared for that conventional companion designs cannot.
  • Building companions is hard; the Elizabeth pattern is one of the few known paths that reliably works at AAA scale.
  • If you must build a companion for your own game, start from the Elizabeth principles and loosen only where your design specifically benefits from tactical presence or risk.