Development of Deus Ex

The original 1997 design document for the game privileges character development over all other parts, including experimental sequences and technology demos.[8] They wanted players to consider "who they wanted to be" in the game, and for that to be connected to how they behaved in the game.[4] In this way, the game world was "deeply simulated": real and common sense enough to be believed that the player may think about solving the game problems in creative and emergent ways without seeing distinct puzzles.[4] The developers also wanted "choice" with "consequence"—what Spector called the team's "two most frequently uttered words".[4] Their simulation was not able to maintain that level of openness, and the team had to force "skill", "action", and "character interaction" paths through the levels.[17] Spector compared this technique to the practices of Ultima's developers, though he felt his team did it "more consciously and ... effectively".[17] Spector later credited Konami's 1995 role-playing video game Suikoden as an inspiration, stating that the limited choices in Suikoden inspired him to expand on the idea with more meaningful choices in Deux Ex.[18]

The game changed greatly over the course of production, but parts that remained consistent include the augmented counterterrorist protagonist JC Denton. The anti-terrorism organization UNATCO was originally TLC, the Terrorist Limitation Coalition. Ally Tracer Tong was more of a "mercenary" than a "kindly anarchist", while enemies like UNATCO's Joseph Manderley went from "ruthless bastard" to "stuffy bureaucrat" and Majestic 12's Bob Page and assassin Anna Navarre played more of a background role.[8] Many of the original character ideas were reshaped to fit the final game design. The Majestic 12 organization originally intended to initiate a Mexican invasion of Texas and then suffocate the presidential cabinet by killing their oxygen supply. When this failed, their artificial intelligence kills the organization and retreats to outer space with nuclear weapons. Some of the plot and characters were brought to the final version, except that Majestic 12 is clandestine and focused on controlling the Internet. Though Spector originally pictured the game as akin to The X-Files, lead writer Sheldon Pacotti felt it ended up more like James Bond.[8] Spector himself called the game "James Bond meets The X-Files".[6] He wrote that the team overextended itself by planning such elaborate scenes, especially parts such as a replica of downtown Austin,[12] Area 51 reconstructed from satellite data,[17] a sunken post-earthquake Los Angeles, a raid to free thousands of prisoners of war from a Federal Emergency Management Agency-controlled United Nations concentration camp, and over 25 missions throughout Siberia, western Europe, and the United States.[17]

Designer Harvey Smith

Designer Harvey Smith suggested a streamlined plot that removed the Mexican invasion to make development easier and the narrative more personal. While finished assets were repurposed, entire parts of the game were abandoned, including Texas and the Denver airport. Environments such as a flooded city became a Majestic 12 research site and the endgame Helios space station became Area 51. He also made the call to remove a playable mid-game White House level due to its complexity and production needs in other areas. Pete Davison of USgamer referred to the White House and presidential bunker as "the truly deleted scenes of Deus Ex's lost levels".[8] Spector referred to his "dream" level[17]—the White House—as their "toughest map challenge".[19] The team has speculated that the assets may exist on a "DVDs in someone's attic", though Pacotti, Smith, Spector, and the final release have no trace of them.[8] Pacotti felt that the experience of rescuing Paul was worth cutting the White House from the game.[8]

Testing revealed that their idea of a role-playing game based in the real world was more interesting in theory than actuality.[19] They chose two real-world inspirations for levels: "highly interconnected, multi-level" spaces and places one cannot normally visit (such as the White House).[4] In practice, they felt that some functions of the real world, such as hotels and office buildings, were not compelling in a game,[19] and that "reality" always lost when up against "fun".[4] Their recreations of notable locations and items such as the Statue of Liberty and payphones were questioned when they did not emulate the actual site or function. Internally, the team began to doubt their investment in a non-player character-driven game as sufficiently interesting. Spector was swayed by this widespread sentiment to have "monsters and bad guys", and the team increased the prominence of several robots and added genetically-altered animals that still fit the story.[19]

The game was designed to be "genre-busting"—partly simulation, role-playing game, first-person shooter, and adventure.[6] Spector intended for Deus Ex to be an "immersive simulation" similar to Ultima Underworld[5] by removing reminders of the game world such as interface or backstory.[6] He described the game's role-playing elements as the player-character becomes a "unique alter ego" built from the ramifications of their unique gameplay decisions.[6] Similar to a first-person shooter, the game uses a first-person perspective and includes shooting, but there are other non-violent gameplay options. Like an adventure game, Deux Ex is primarily based on a linear narrative story and item collection, though its puzzles are open-ended with many possible solutions and consequences. Ion Storm felt the game was about player expression rather than the appearance of the developers' ingenuity, and wanted to treat the player as a "collaborator" that they could empower to make choices and deal with consequences.[6]

Once implemented, the team's game systems did not work as intended. They built prototypes of the systems and some missions towards the beginning of development, which uncovered some of their blind spots.[10] For example, their early implementation of the conversation system and user interface afforded them enough time to revise by the final release.[12] They also found the augmentations and skills to not be as interesting in implementation as they seemed in the design document.[10] Developers from other companies, such as Doug Church, Rob Fermier, Marc LeBlanc, and Gabe Newell, identified these deficiencies in game "tension" when they played the prototype.[19] This led to a substantial revision of the augmentations and skills; Smith designed an augmentation per-use "energy cost" that led to energy refill rewards and more tactical play.[17]

Their milestones served as wake-up calls for the game's direction. A May 1998 "proto-mission" milestone was to have a basic but functional demo of the crucial game systems and two maps (the White House and Hong Kong) ready.[19] The team worked on the riskiest parts of the game first such that the core game would be functioning, albeit not polished. The demo revealed that the size of their maps caused performance problems and would have to be divided into smaller maps. It also was one of the first signs that maps needed to be cut. The team forgot to prototype the non-player character artificial intelligence in this demo. A year later, in May 1999, they reached a milestone for finished game systems and the first two missions completed. While the player could start the game and character, use augmentations, acquire inventory, complete objectives, and save the game, their poor quality and comments from Newell led Spector to nickname the milestone the "Wow, these missions suck" milestone.[19] They felt that their first demos showed their potential, but they were not nearly close to reaching it. This milestone also helped them estimate the work required for missions and the portions that worked best, which led to a trim of their 500-page design manual to the best 270 pages.[19] Spector recalled Smith's mantra here: "less is more". [20]

Smith had "an intimacy" with the level design tools such that he could sense when the design met the game's technical restrictions, like the maximum possible size or number of characters for a room. He suggested the narrower skill tree with palpable effects, such as exchanging skills about weapon damage for more accessible weapons.[8] The team adopted the cuts, having been encouraged by the recent milestone. Spector has said that had they waited for the beta to have made the same reductions (as per common practice), "it would have been a disaster."[21] Spector's post-proto-mission strategy was to bring the rest of the missions to proto-mission functionality rather than perfecting the two existing proto-missions. He felt that this would be a more efficient use of staff time, even if it meant that the final product would be more "bare-bones" than brilliant.[16] Almost all of their game systems besides the inventory and conversation schemes were rethought up through the end of the project.[22]

They also made choices about the minutiae of game mechanics. The team chose to pause the action while the player was viewing interfaces as to keep the strategy tactical, and to make the affinity of strangers and purpose of items "instantly recognizable".[22] Spector said that the issue of naming the player-character almost incited a "holy war", though they compromised and let the player choose in exchange for a common "code name and backstory".[22] They also saved the cinematic cutscene work for last so as not to repeat work as things changed during development.[22]

Spector wrote that the team did not figure out how to handle NPC AI until late in development.[12] This led to wasted time as the team discarded their early AI. They were also building atop their game engine's shooter-based AI instead of building a customized version anew to handle the spectrum of convincing emotions they wanted from their characters. As a result, NPC behavior was variable until the very end of development. Spector felt that their "sin" was their inconsistent display of a trustable "human AI".[12]

Deus Ex used the Unreal game engine, which was designed for first-person shooters.

The game was developed on systems including dual-processor Pentium Pro 200s and Athlon 800s with eight and nine-gigabyte hard drives, some using SCSI. The team used "more than 100 video cards" throughout development.[6] Deus Ex was built using Visual Studio, Lightwave, and Lotus Notes. The team also built a custom dialogue editor, ConEdit.[6] They used UnrealEd atop the Unreal game engine for map design, which Spector wrote was "superior to anything else available".[21] Their trust in UnrealScript led them to code "special cases" for their immediate mission needs instead of more generalized multi-case code.[12] Even as team members expressed concern, the team only addressed this later in the project. To Spector, this was a lesson to always prefer "general solutions" over "special casing", such that the tool set works predictably.[12]

They waited to license a game engine until after preproduction,[1] expecting the benefits of licensing to be more time for the content and gameplay. They chose the Unreal engine as it did 80% of what they needed from an engine and was more economical than building from scratch. Their small programming team allowed for a larger design group. The programmers also found the engine accommodating,[21] though it took about nine months to acclimate to the software. Spector estimated that six to nine months of playing with the engine were necessary to learn how to use it properly, and even still, their ideas did not always implement as planned.[10] Spector thought it would have been much harder to have written the interfaces, skill systems, and conversations on their own. Despite the savings, the time Spector thought they would save not writing an engine was lost learning the engine, though they did have more time to work on content creation and gameplay systems. Spector also felt that they would have understood the code better had they built it themselves, instead of "treating the engine as a black box" and coding conservatively.[21] He acknowledged that this precipitated into the Direct3D issues in their final release, which slipped through their quality assurance testing.[21] Spector also noted that the artificial intelligence, pathfinding, and sound propagation were designed for shooters and should have been rewritten from scratch instead of relying on the engine. He thought the licensed engine worked well enough that he expected to use the same for the game's sequel and Thief 3.[7] He added that developers should not attempt to force their technology to perform in ways it was not intended, and should find a balance between perfection and pragmatism.[12]