Don't Try to Pen Them In: Todd Howard on Skyrim, Starfield, and Accidental Doors

Lex Fridman opens with a half-joke that sounds like philosophy homework: if we are living inside a future Todd Howard game, how would a character escape? Howard does not wave it off. The games Bethesda Game Studios makes are supposed to be as open as possible, so the first thing testers do is establish the rules. They try to pick up every mug, try every door, collide with every wall, and ask what the world will actually allow.

That stance — say yes as much as possible — produces chaos. If you were stuck in one of those worlds, Howard says, you would try everything, and you would usually find a door or a stack of crates the designers never expected you to pile against a wall. Fridman names it: not a designed doorway out, an accidental one, a happy bug. You could do a *Truman Show* and just keep going. Howard agrees. Just keep going.

The catch is that realism and scale make the seam harder to find, and players get more patient about searching. The studio learned the practical lesson: they are going to find a way, so do not try to pen them in. Bethesda even leaves a developer test cell it does not expect anyone to reach, usually stacked with crates of every weapon in the game. Players find it anyway.

Skyrim's guards became a meme factory because they clock what you are carrying and what you just did. Howard's first rule for NPC dialogue is not backstory. It is reactivity. Characters written for film need to be interesting for themselves; some game characters work that way too, and players do care about them. The ones that stick are the ones that talk back to *you*.

Guards are written almost purely to react: nice jacket, cool watch, hey, what did you do. Lydia, the housecarl everyone quotes, says "I'm sworn to carry your burdens." Howard notes it is a generic line every housecarl has. It lands because of a slight snark in that particular read, and because you are asking her to do something. She is answering you, not reciting a biography.

The technical version is modest. Dialogue lives in stacks that behave like a small state machine: here is what is happening, here is a random list of things I could say. The team hunts for player behaviors the NPCs can catch — jumping on tables, stealing, sneaking in at night — because the more of those they can answer, the more alive the characters feel. Underneath, he thinks of the AI as one big manager for everybody, not a unique brain per person.

Fridman wants to know how large those state machines get. Howard's answer is more centralized than the question assumes. There is a manager for all the people. What makes Bethesda's worlds unusual, he says, is that they spend cycles thinking about everybody in the world all the time. Distant NPCs drop to a much lower tick rate, but the studio still knows if someone wants to walk across the map, and it runs every quest at once.

Other open-world games start an activity and shut the rest of the world down so that moment can feel important. Howard would rather keep the rest of it going. It is more of a simulation, and the interesting scenes happen when systems collide: you are on a quest, an NPC arrives on their own schedule, and nobody wrote the overlap. Close NPCs simply update more often. The trade is compute, which is one reason the studio has historically accepted 30 frames a second.

Fridman pushes further: can the town have a history while you are away? Howard is blunt. They do not prioritize what you cannot see. Level designers studied how Disney parks are built — shops in front, a planned first impression — because the world still exists *for the player*. The little things, wind and a line of dialogue and who is sitting in the tavern, are there so you buy the virtual reality. Then a dragon landing or a deathclaw walking out of the wasteland can hit you the way it would hit the people in that world. Looking ahead, he wants factions designed by how they exist first, not as a storefront where you sign up, do quests, get gold, and become guildmaster.

Fridman's first memory of *The Elder Scrolls II: Daggerfall* is immensity — the sense, when you are young, that the future is wide open. Howard says the 1996 game still feels endless even though the map has edges, because of how much physical space they gave villages and towns around the Iliac Bay. *Arena* (1994) covered all the provinces and followed a similar pattern, but it did not have Daggerfall's count of places. Scale on the map and density on the ground are different variables.

Daggerfall became a touchstone for *Starfield*. Wander outside a city in Daggerfall and the terrain is generated; wander Skyrim and every corner was placed by hand. Starfield is closer to the older game: leave the city and a lot of what you meet is generated. Howard admits it is a game flow Bethesda has not really shipped before, and it is fun in a different way. *Arena* was the elegant XP loop — do this, get experience, level up. Daggerfall dug into who your character is, skills, advantages, disadvantages, and a full 3D world you could look up into, instead of Arena's Doom-style 2.5D.

You could walk outside the city in Arena too, Howard says, but it looked fakey: a flat plane, things coming at you, a dungeon entrance as an 8-bit flap toward the camera. Fridman remembered Arena as wide open. Howard's correction is gentle and useful. The feeling of no edge belonged to Daggerfall. Sense of scale is not a function of how many provinces you painted on a map. It is how much you bump into when you walk.

Howard's first games were arcade machines at the pizza place: *Space Invaders*, then *Pac-Man*, which he credits with popularizing video games at a scale that felt insane — a song, a cartoon, merch, the whole apparatus. The hook that made him a maker arrived in fifth grade, when a TRS-80 came into the classroom with a *Star Trek* game on it. Teachers asked if he wanted to know how it was made. He was hooked. He needed to figure out how to make this stuff. By sixth and seventh grade he had a better *Star Trek* clone on an Apple II.

Games, including Dungeons & Dragons on paper and *Wizardry* and *Ultima* on a computer, were a way to experience things he could not. He made a *Raiders of the Lost Ark* game, a *Miami Vice* game, a *Groo the Wanderer* game, a *Traveller* game, each time chasing a new Apple II graphics trick — editing character sets, enabling the secret double-high-res mode. Friends were doing the same. He puts himself middle of the pack. He did not, in the slightest, think this would be his life. He did not even think it was a job. College computer classes were not where he wanted them, so he took a business degree and interviewed for finance.

He reached the final round of a corporate finance job at Circuit City and they turned him down. His reaction, as he tells it: forget them, he was going to make video games. His girlfriend, now his wife, had given him *Wayne Gretzky Hockey 3* for Christmas. The box listed Rockville, Maryland, which sat on his drive back to the College of William & Mary. On Martin Luther King Day in 1993 he drove over and knocked. Someone came out and told him to contact them when he was out of school. He kept calling. *Arena* had just come out and he loved it. Bethesda was still a sports house — *Gridiron!*, hockey, NCAA basketball — and there was an industry story that Electronic Arts, struggling with *Madden*, had hired Bethesda for help. When Howard started, the owner's office held John Madden's Oakland Raiders playbook. He still plays *Madden*. He wanted to make the ultimate college football game. What made him stay was that the company was also doing Terminator science fiction and open-world RPGs. That, he says, was where he had to work. *Arena* shipped on something like six or eight floppy disks, and disk count was money. CD-ROM was the industry's jumping-off point because the data finally felt unlimited.

Howard's first assignments were finishing *NCAA Basketball: Road to the Final Four 2* and *Terminator: Future Shock*, which was in production beside *Daggerfall* on a new 3D engine, XnGine, written by programmers out of the Danish PC demo scene. Bethesda development was maybe twenty people. Daggerfall sat a bit behind, so he became a mainstay on Terminator, quickly a producer, making levels. What he did not appreciate at the time was that nobody quite owned the Terminator license. It was in legal limbo, so nobody could tell them no. They picked the movies apart — which gun, the wattage of the laser.

Future Shock is a large open post-apocalyptic landscape, a height map with instanced objects all over it, and Howard says that is still a lot of how they build worlds. Instancing means you do not store every wall as unique data. You build husks of buildings and place them everywhere, which is kinder to memory and rendering and lets you make a bigger world faster. In those years the technology moved every season: new methods, 3D acceleration, HD. Visual returns have diminished. You can now spend a fortune on the roughness of a cup and how global illumination bounces off a wall. You can pretty much do what you want if you put the time in. Then, you could not, so you picked your battles.

Asked whether graphics now trade off against story, Howard rearranges the question. They start by giving the player as much agency as possible, then sacrifice some fidelity and some speed. Traditionally the games are okay at 30 frames a second as long as they look good and the simulation is running. Art time per cup has exploded, so they outsource like everyone else, and relative to the industry they still run smaller teams. The awe of stepping out of a Skyrim inn is graphics plus mood, which he discusses as tone. He likes the quiet: how the wind sounds, how trees and clouds move, strolling to watch the sunset land on water. The sky takes more of the screen than people credit. Soundscape matters as much as pixels. On Xbox 360-era Skyrim, Oblivion, and Fallout, a huge share of the work was streaming data as you walked and simplifying distant geometry so the processor survived. Today's drives make that ordinary. Then it was the hard part.

Tone, for Howard, is all of it together. Flip through a few images and you can tell *The Lord of the Rings* from *Game of Thrones* from *Excalibur* from the Syfy series of the month. He looks for what will make a setting unique, then leans on something grounded in reality and lets the more fantastical pieces build. *Starfield* has laser guns and ships that shoot each other, but you should be able to look at it and say this is an extension of how we view space today. *Fallout*, he admits, gets a little crazier the longer they stay in it.

Even a fantasy world has to answer practical questions: how prevalent is magic, are weapons for utility or decoration, how do people live. The test is whether the place feels welcoming because it is nailed to a reality you understand, historical or near-future. *Morrowind* gets weird — mushroom trees, giant insects — but it intentionally starts in a classic medieval-looking town, and you look just beyond the walls to see the strangeness. Dragons in Skyrim were debated as tone, not as a meme. Howard wanted beasts that could exist in that world, less purely fantastical, introduced on a slow roll so NPCs are reacting to sightings the way the player is. *Daggerfall*, to his memory, had dragon wings more than dragons. *Redguard*, his adventure game, brought a dragon back.

Fridman loved *Redguard* and asks why it flopped. Howard does not flinch. Coming off Daggerfall, with Morrowind in design, the team split; *Battlespire* and Redguard went their ways. He wanted a handcrafted, Ultima-feeling blend, a love letter to *Prince of Persia* with some *Raiders* in it, and it would have lived better on a PlayStation or the Xbox that did not exist yet. Tomb Raider had just arrived. 3D cards were new. It was one of the last DOS games in a Windows world, so it missed a technology window, and it was not what people wanted from Bethesda. The company almost went out of business, then reformed under ZeniMax Media with Robert Altman. Howard felt personally responsible. The lesson he took was not to doubt that Redguard was a good game. Critically it did well and won awards. It did not sell. His takeaway: he was not going to be conservative again. Swing for the fences, accept rough edges, rather than ship something timid.

A Reddit AMA leftover: during Redguard, the lead programmer made all the buildings hop up and down after ten minutes just to mess with Howard. The serious version of that story is that open-world programming is about robust systems, because players stay for a very long time and you cannot touch everything by hand. When two systems meet, they should not explode. People confuse engines with renderers. Most of the engine is how you bring data to the thing that draws pixels: walking the world, feeding geometry, drawing it. Then physics and interactivity — which objects are scenery, which ones you can sleep on, sit in, pick up, use to bake bread. Bethesda puts a premium on the touchable ones.

AI loops back to that people-manager, with combat on top. One of Howard's favorite moments is an NPC scanning the environment, spotting a better weapon than the one they hold, and using it on you. In *Oblivion* they learned the hard way not to let enemies loot the designer's treasure: they would enter a level and find every foe armed to the teeth and nothing left for the player. The rule became that NPCs take weapons dropped by other NPCs or the player, not the chest you were supposed to open. Looking forward, he wants more reactivity and roles in the world so characters are not just signposts. On language models, he has seen incredible demos. He thinks the tech is coming for the kinds of games they make. The open questions are what is ready to ship, whether you generate on the fly or bake static lines, and whether voice runs on the client, a server, or in the build. They already record voice for three or four years across a project, starting early, because a great actor makes the writing better or ad-libs it into something the page did not have.

Players fall in love with companions. Howard does too, in other people's games. *Starfield* pushes relationships a step past a like-you meter: a companion can be in love with you, hate something you just did, be angry for a while, and come back. He does not want to oversell the complexity, but he wants the gray area instead of a romance that un-happens when you drift out of a state. The time people spend in games is real time to him. Achievements and libraries are almost a scrapbook. He wishes he had that list back to the late 1970s. Kids today will.

The studio had always wanted an explore-space RPG in the Bethesda style. Howard loved the pen-and-paper game *Traveller* (one of the first things he tried to make on the Apple II), plus *Starflight*, *Star Control II*, and the lesser-known Apple II title *SunDog*. After bouncing between Elder Scrolls and Fallout, it felt like time. Fridman notes the leap: those series live on one planet; this one talks about on the order of 100 star systems and 1,000 planets. The first technical question after *Fallout 4*, starting in 2016, was how to render a planet you could actually land on, instead of a handful of crafted pads. Tone said you should be able to land on that ice ball. Fractal landscapes looked like muddy goop. They prototyped large landscape tiles the way they usually build terrain, generated and hand-touched offline, then wrapped them around a sphere and blended them. It looked reasonable. They thought they could do it.

The design problem was sharper: what is fun about landing where there is potentially nothing? A lot of real moons are resources and loneliness. They leaned into that, as long as they tell the player what is there. Howard equates it to listening to the wind and watching the sunset in Skyrim. There is a beauty to being the only person on a strange planet, building an outpost, watching a gas giant rotate through sunrise because the system is actually modeled. Most people will not notice every orbit. They still get to say they want to see that on that place, while quests point at the handcrafted content. Once you can build a planet, 100 or 1,000 or a million is changing a number. What you cannot do at huge counts is name them, validate them, and know how a far system feels. They leveled star systems the way a map marks low-level zones, and that control is why they landed around a hundred systems.

Space travel is "dangerous" mostly as tone. They used to strand you when you ran out of fuel; on paper it was a great beacon-and-rescue moment, in play it stopped the game. Environmental hazards and suit buffs stayed. Robots are in, including the companion Vasco, but they dialed generic utility robots back because a chart of the real future would give machines a bigger role than this human-centered tone wanted. Visiting SpaceX felt to Howard like the Avengers meeting NASA. He is in awe of the engineering at the edge of physics, and of watching a SpaceX launch from the Delaware coast and thinking, for a stunned minute, that it was a UFO. Science fiction, he hopes, might spark a young player toward engineering. He is encouraged by talk that the James Webb Space Telescope could find signs of life in a couple of decades, including industrial byproduct gases that would imply a society. Asked for an Elder Scrolls VI date, he will not blink. He has a vague idea. He wishes it were soon. It will take as long as it takes, and Skyrim, eleven years old at the time of the recording, is still probably their most played game. People may still be in it a decade later, which changes how you architect the next one if they will play that for twenty years.

Howard has spent more time in Tamriel than anywhere else in his life, and when they are not making an Elder Scrolls game they miss it. Yesterday's meeting, he says, left him wanting to play the next one right now. They will make sure they do it right. Creating a world that stays moddable starts inside the studio: everything they do, a content creator will have to do with the same tools. They clean the kludge for release. People have been modding Bethesda games for twenty years. The Creation Kit on Steam lets you change the weather color and ship a tiny plugin, or spend years on a quest line, which is more than a hobby. They have hired a large number of modders as professionals and want others to be able to make a career of it without joining. Pushing AI usually means getting into code, which some people do with script extenders; a new game simply gives the community a bigger palette.

Internally they already run very dumb bots every day, loading every space in Starfield, checking crashes, memory growth, and frame rate, plus a heavily scripted bot that plays the first main quests on every build so QA is not burned in five minutes. Fridman wonders about a boredom meter. Howard's working chart is simpler: games get put down from frustration and confusion or from boredom, and you want the player in the middle. *Dark Souls* is the case where people leave angry and come back stronger. Skyrim's survival mode — hunger, heat and cold — was a Creation Club piece by an external creator who is now full-time.

A Bethesda game usually starts as lunch conversation: tone, setting, a big-ticket feature that makes it unique. By the midpoint of one title they know the next. They concept art and music early; they already did a take of Elder Scrolls VI's music for the 2018 teaser and have taken it further. Howard likes to have the beginning of the game worked out — what the player does first. Then a first playable, a vertical slice with part of the team, full production for a year or two or more, and a final six months to a year of glue so features do not feel like different games. The skill system and HUD are the party host, pacing activities until something called game flow exists, which it does not until that last year. Combat has to feel good in the hand, with a little magnetism players do not notice, and enemies who perform the illusion of being smart while mostly being there to be killed. Fully smart enemies are not fun. "Let the wookiee win." Crafting clicked for him in *Fallout 4*, where everything in the world could be an alchemical component, so a cup was not just one cap of junk.

Howard makes the case for *Fallout 4* as a different game with a different tone from *Fallout 3*. What makes Fallout tick is not generic apocalypse. It is the world that blew itself up: a 1950s American dream of nuclear utopianism gone wrong, drama and darkness woven with B-movie humor that winks at the camera, duck-and-cover irony. When Bethesda announced Fallout 3 they had finished Morrowind but not announced Oblivion, and they expected to anger Elder Scrolls fans and hardcore Fallout fans at once. The original was a beloved isometric turn-based RPG. Two and a half years after Oblivion, plus DLC, Fallout 3 found its audience. It is one of Howard's favorite projects because everything felt new. He still loves the beginning: you are born in the Vault, raised there, and when you step out you need to find your father. Skyrim leaves you a blank slate. Fallout is stricter about who you were. Starfield splits the difference with a chosen background.

Over time they have tried to stop players from restarting because they built the wrong character eight hours in. Post-Oblivion they moved toward classless systems. In Skyrim you pick a race and some starting skills; almost everything else you can grow into, the way you might decide tomorrow to learn archery. Visual appearance can change later. Spellmaking got guardrails after Morrowind because recall stones teleported people out of quests. Designers would build a quest, someone would say "and then I recall away," and one day the room agreed to cut the effect so they could make better content. Howard's favorite quests are not always the handcrafted story with a twist. Sometimes they are "find me five daedra hearts," objects placed believably in the world, with the how left to the player.

A clutter group treats interiors like treasure-and-trash decorating. Too many decision-point T-intersections and players get lost; if they check the map, Howard feels the level has somewhat failed. Every game they under-clutter, then over-clutter, then declutter, because too many important objects in a room make you feel like an idiot if you do not click them all. Rare drops are an art they are still learning from *Destiny* and *Diablo*. They design "did you know" moments for the water-cooler, because their games have classically been single-player. *Fallout 76* and *Elder Scrolls Online* are the exceptions; ESO's storytelling, he says, is excellent. Mobile is a different onboarding problem because the game is free and sessions are short, though *Fallout Shelter* players will sit for hours. Internal deadlines are best-guess honesty, not arbitrary crunches. They announced a Starfield date and delayed it. The risk to fans, the game, the team, and Xbox was too high. Microsoft's $7.5 billion purchase of Bethesda and ZeniMax felt, from the inside, like the player-first culture people see from the outside. Starfield is an Xbox exclusive, which is not strange to a studio that led on Xbox from Morrowind onward. Howard likes knowing the hardware he is writing for, and the Series X with Xbox's own engineers is, from his seat, awesome.

Asked for the greatest game of all time, Howard says *Tetris*: elegant, a challenge anyone can enjoy, a thing that only works because it is a video game. Fridman puts Skyrim first; Howard will take it. Off his own catalog he names *Ultima VII* as a personal landmark, a virtual fantasy world where you could break bread and pick everything up, each entry with its own tone and a cardboard box that felt like an object. He still cares about cover art lining up on a shelf, even digitally. *Pac-Man* brought games into the mainstream. Nintendo still looks to him like the best game-makers in the world. *The Legend of Zelda: Breath of the Wild* is his favorite Zelda because it constrains you with abilities you have not earned, not arbitrary barriers. *Grand Theft Auto III* ushered in 3D open world on the PlayStation 2; *Vice City* might be his favorite of that series; *Red Dead Redemption* might be his favorite story. He could go on: *Portal*, *Hearthstone*, *NCAA Football* as a role-playing game with a roster that graduates, *Forza*. Recreating a real sport everyone already knows, every year, is a harder job than people think. He buys *Madden* annually.

A good day is not inbox zero. He executive-produces, directs Starfield, glances at the Fallout TV show and the Indiana Jones game at MachineGames, then plays some of Starfield, looks at the problem set, and sits with other developers until they solve it with the pieces in front of them. Pie-in-the-sky is fun. Fixing the build is the reward. The studio is programming, art, quest and systems and level design, writing, production, audio, QA. It is, he says, the greatest job you could have. They now get data on where players go, die, and drop off. The emotion he likes to name is pride — look what I did — plus the escape of being more powerful than you are in life. Make-A-Wish visits, kids who could ask for anything and want to see the next game, remind the team how many hours of happiness a tiny decision can create. Advice for the young: find something you love so much it never feels like a job, do not do it for money, never underestimate yourself, stay resilient because you love it. He slept on a sofa when he started at Bethesda. When the company almost died after Redguard, he treated it as a lesson. *Fallout 76*'s bad launch became a thing to edit until it found success.

After Fallout 3, movie producers wanted a two-hour synthesis of the game. He stalled. Prestige television changed the math. Jonathan Nolan and Lisa Joy, of *Westworld*, took it on; Nolan is executive producer and directed the first episodes. Howard's joy was that Nolan already loved Fallout 3. The series is not a retelling of Fallout 3 or 4. It is a new story in an area of the map that does not break the rules, so people who have never played can still feel how strange the world is. TV, he noticed, is more character-driven; the NPCs have to do the work because you cannot walk in. Indiana Jones is the opposite problem: the character *is* the world. *Raiders of the Lost Ark* is still his favorite movie, the temple and the boulder the greatest opening, a story he believed as a child. He pitched Lucasfilm an Indiana Jones game in 2009; the deal fell apart and they made Skyrim instead. A decade later Disney was licensing, MachineGames of *Wolfenstein* was the right storyteller, and the result is a mashup, a love letter, not one genre. He cannot imagine doing anything else, but after nearly thirty years he wants the team to enjoy development itself. Most of his life is Tamriel and the wasteland. They do not get that many shots. If life is a video game, what motivates him is curiosity — not vacuuming up facts, but noticing how other humans tick. He quotes the line that everybody has two lives, and the second starts when you realize there is only one. Skyrim's tenth anniversary let him see the game through his younger son's obsession, lore at dinner, mods and YouTubers, then four hours he had to force himself to turn off.