DESIGNER-STUDY · 2026-06-27

Inside Maddy Thorson's Philosophy — Keep It Hard, Make It Kind

TowerFall / Celeste, and designing invisible kindness

Introduction

Maddy Thorson is a Canadian indie developer best known for the four-player local-multiplayer archery game TowerFall (2013) and the hard-as-nails mountain-climbing platformer Celeste (2018). Celeste is famous for being a brutal game that nonetheless feels kind, and its accessibility design — including the optional Assist Mode — is discussed endlessly. She is not a puzzle designer per se, but her practice of building challenges one screen at a time, and of reconciling difficulty with kindness, is well worth studying for anyone thinking about puzzle design.

What draws me (Kizuki) to her is that she keeps difficulty and kindness in the same room rather than treating them as opposites. Across interviews and blog posts, her obsession consistently points one way: quietly helping the player. This piece is not a walkthrough; it is a study of Maddy Thorson the designer — her philosophy, obsessions, failures, dilemmas, and influences — built from her own words. (Sources from before 2018 use her former name; throughout, I refer to her as Maddy and she.)

Background — from freeware to the mountain

Thorson spent years making freeware and Flash games — Jumper, An Untitled Story, Give Up, Robot — before her commercial debut, TowerFall. TowerFall launched in 2013 on the Android-based microconsole Ouya, then expanded to PC and PS4 as TowerFall Ascension, gained the "Dark World" expansion, and reached the Switch (Rolling Stone, 2024). She calls the pared-down Switch edition the definitive one.

Her next game, Celeste (2018), co-created with Noel Berry, landed on nearly every game-of-the-year list. Its story — Madeline climbing a mountain while confronting her own anxiety — came to mirror the developer's own. Thorson came out publicly as a trans woman in 2020, and Rolling Stone reports she has been open about the role Celeste played "in discovering herself as a woman" (Rolling Stone, 2024). She now leads the studio Extremely OK Games (EXOK).

Philosophy — keep it hard, make it kind

The most consistent thread across her statements is this: keep the difficulty, but make it kind. On Celeste's Assist Mode she says, "In Celeste, we try to gently push the player to do things that they thought was impossible for them" (Vice, 2018). Difficulty, in her framing, is not punishment but a device for becoming capable.

The goal is explicit too: "We want people to come out of this game feeling capable and powerful, so that means we have to teach them, challenge them, and support them through the failures along the way" (Vice, 2018). Challenge and support are not separate; she designs them as one continuous gesture, and that, I think, is the core of her work.

Obsessions — tilt everything a little toward the player

A recurring fingerprint in her work is a habit of fudging the controls "just a little" in the player's favor. In her own write-up, Celeste & Forgiveness, she lists coyote time (you can still jump shortly after leaving a ledge), jump input buffering, corner correction when you bonk your head, widened wall-jump windows, stamina refunds, and more, concluding: "All are centered around widening timing/positioning windows, so that everything is fudged a tiny bit in the player's favor" (Celeste & Forgiveness, 2020).

And: "I think this is a big reason why Celeste can feel kind even though it's very difficult — it wants you to succeed" (Celeste & Forgiveness, 2020). In the same piece she adds, "I don't think we invented any of these" — no posturing as an inventor, just a craftsperson carefully stacking known techniques.

Another obsession is paring verbs down. TowerFall is only shoot, jump, dash; Celeste reduces to move, jump, dash, climb. On making every TowerFall character identical, she says, "I felt like I had to make it as simple as possible" (Rolling Stone, 2024). Deep situations from a small vocabulary — a way of thinking that maps cleanly onto puzzle design.

Failures and recovery — choosing the fun thing instead

The biggest pivot she discusses publicly is TowerFall itself: "At the very start, it was going to be single-player ... but it just wasn't working. And then I prototyped the multiplayer and, immediately, that was fun ... I was like 'screw the single-player'" (Rolling Stone, 2024).

Going multiplayer was a commercially scary choice. "It seemed like a really dumb idea, but it wasn't up to me. [That] was what was fun about the game, and it felt like a waste to shelve it" (Rolling Stone, 2024). She followed where the fun lived rather than what was easy to sell — quick to abandon a failing premise and move to what actually worked.

Another is naming Assist Mode. It was first called "Cheat Mode," which felt "judgmental." "That judgmental feeling really was what drove us to continue workshopping the name ... we eventually landed on Assist Mode" (Vice, 2018). Realizing how her own words made players feel, she reworked the name and the explanatory text many times — a small failure, self-corrected.

Dilemmas — authorial intent, and letting go

The dilemma she voices most clearly is the clash between authorial intent and player freedom. On Assist Mode: "From my perspective as the game's designer, Assist Mode breaks the game. I spent many hours fine-tuning the difficulty of Celeste, so it's easy for me to feel precious about my designs. But ultimately, we want to empower the player and give them a good experience, and sometimes that means letting go" (Vice, 2018).

She then kept searching for non-coercive language: "The goal is a fluid experience where players are safe to float around between loosely-defined difficulty levels as suits them, without judgement or implication that they aren't playing the game 'as intended'" (Vice, 2018). Keeping the difficulty boundaries loose and never implying a player is off the "intended" path — between intent and freedom, she chose not-judging as her answer.

A second tension is total immersion versus sustainability. Recalling her early development, she says the thought "I just have to finish this game, and then I can die" went through her mind repeatedly, and now calls it "terrible." Yet she also says "there's something nice about that, even though it was not a sustainable lifestyle," holding both the dignity and the harm of that devotion at once (Rolling Stone, 2024).

Influences — Nintendo, rock walls, communal living

Her acknowledged influences are concrete. She flatly calls TowerFall a "rip-off" of Super Smash Bros. (Rolling Stone, 2024). Assist Mode grew out of closely watching the conversation around the difficulty-debated Cuphead, an attempt to find a better shape for accessibility (Vice, 2018).

Celeste's screen-by-screen design draws on real-life rock climbing. In her GDC 2017 talk, "Level Design Workshop: Designing 'Celeste'," she discusses climbing-inspired thinking, giving each screen a narrative arc (beginning, tension, climax, resolution), and teaching players implicitly (GDC Vault, 2017 / YouTube).

And one can't overlook life itself. During TowerFall the parties never stopped in her communal-living setup — she even asks, "Did TowerFall make me have all those parties ... or was it the other way around?" — making development and socializing inseparable (Rolling Stone, 2024). Celeste and Madeline, by contrast, are bound to a deeply personal experience of facing herself.

Kizuki's reading

From here, this is my (Kizuki's) interpretation. What runs through Maddy Thorson's design, I read, is a kindness that refuses to announce itself. Coyote time and corner correction are invisible to the player. Assist Mode is offered gently, in non-judging words. As she says, "I don't think we invented any of these" — her authorship is not display but care that works where you can't see it.

And I read TowerFall and Celeste not as opposites but as one impulse pointed in two directions. She herself contrasts the former as something to "bring people together" and the latter as "a poem or a short story," "the main event itself" (Rolling Stone, 2024). Connecting people, and helping one person face themselves — both are the same care: design that empowers. One contradiction is worth naming. She built Celeste, a game about self-forgiveness, by pushing herself so hard she thought "I just have to finish this game, and then I can die" (Rolling Stone, 2024). In the gap between what the work preaches and how it was made, I see this designer's urgency.

Closing — where to start

To understand her, play a few Celeste screens without Assist Mode, then read her blog post Celeste & Forgiveness. You'll feel the resistance first, then the invisible kindness working behind it. And if friends are over, put on TowerFall — it shows the same author's people-connecting face.

On the theme of pairing difficulty with kindness, other designers we've covered serve as useful contrasts: Stephen Lavelle, who makes not-explaining part of the work; Alan Hazelden, who named "thinking"; Jeppe Carlsen, who builds on trust. Reading Thorson against them sharpens the outline of her particular non-judging kindness.

Sources

Primary sources referenced in this article:

GDC 2017: Level Design Workshop — Designing 'Celeste' (her own talk) / YouTube

Vice (Waypoint) — Why The Very Hard 'Celeste' is Perfectly Fine With You Breaking Its Rules (email interview), Feb 2018

Her blog, "Celeste & Forgiveness" (originally a 2020-03-12 Twitter thread; archived on her site) / Medium

Rolling Stone — Why Maddy Thorson Could Never Make a Sequel to 'TowerFall' (interview), Mar 2024

Reactions (no login)

Anonymous • one of each per visitor per day

Read next

FEATURED ESSAY · 2026-06-27

"We didn't want to make just another shooter": Capcom's Pragmata on the puzzle-shooter, and Draknek on what a puzzle game is

Two pieces today. First, I read in the original English a design feature on Game Developer (formerly Gamasutra) about Capcom's new third-person shooter Pragmata, a rare "puzzle shooter" that layers a real-time, Snake-style hacking puzzle on top of combat. Its leads (director Cho Yonghee, producers Naoto Oyama and Edvin Edso) say they "didn't want to make just another shooter," and that fighting repetitiveness while stacking two skillsets was the central design challenge. Second, I read a primary source from Draknek & Friends (Alan Hazelden and co.): their New Voices Puzzle Grant—five $15,000 grants plus mentorship for under-invested-in puzzle designers worldwide—which states plainly what it considers a "puzzle game." From the industry side (Pragmata) and the community side (Draknek), the two together trace the outline of the genre.