The History of Interactive Fiction: And Where AI Takes It Next

Every medium has a moment when it discovers what only it can do. Film found it in editing, when people realised that cutting between two shots could create a meaning that lived in neither, and television found it decades later in serialisation, telling one story over a hundred hours to reach emotional depths no two-hour film could touch. Interactive fiction, I'd argue, is still in the middle of its discovery. The medium has spent about fifty years experimenting with what makes a story work when the reader is the protagonist, and I'm convinced AI is the moment it finally finds its native form.

So this post is a short history of how we got here. I'll look at what each era of interactive fiction made possible and what it ruled out, and then at what AI unlocks that nothing before it could. If you've ever wondered why text adventures boomed and then vanished, or why AI Dungeon felt like a glimpse of something even though it couldn't hold itself together, the arc below traces the thread. It's also, frankly, the story of why I built EchoQuest.

1976: The Colossal Cave

Interactive fiction begins, depending on who you ask, with Colossal Cave Adventure, a text game written by Will Crowther in the mid-1970s. Crowther was a programmer at Bolt, Beranek and Newman (BBN), where he'd helped build the ARPANET, and a keen caver who had helped map parts of Kentucky's Mammoth Cave system. Legend has it he wrote the game for his young daughters. Dennis Jerz, who dug through the original code for a 2007 study in Digital Humanities Quarterly, found that Crowther probably wrote it during the 1975-76 academic year. He also noticed, with a historian's raised eyebrow, that the very last word in its vocabulary table is an expletive, so the kids weren't the only audience Crowther had in mind.

You typed directions and actions, and the game answered with descriptions. A whole world, built out of nothing but words. Don Woods, a Stanford graduate student, later expanded it with fantasy elements (treasure to collect, plus puzzles that needed magic words), and it spread through the early ARPANET like wildfire. One account Jerz quotes estimates that Adventure "set the entire computer industry back two weeks." In 2019 The Strong National Museum of Play inducted it into the World Video Game Hall of Fame, noting that it inspired Scott Adams to found the first commercial computer game company.

Here's my favourite detail, though. A colleague of Crowther's told Jerz that Will was amused by "how well he could fool people into thinking that there was some very complex AI behind the game." Fifty years on, there actually is. I find that delightful.

By any modern measure it was primitive, with a simple parser that knew a short list of commands and a map that never changed. Still, it introduced the core idea everything since has built on. You could stand inside a story and make it move. The story existed somewhere, and you were inside it, changing what happened next with every line you typed. In a sense, every decade of interactive fiction since has been refining that single insight.

The Parser Era (1977–1993)

Infocom turned the cave into an industry. (Fittingly, Infocom co-founder Dave Lebling had played in the same tabletop D&D campaign as Crowther himself.) Zork, The Hitchhiker's Guide to the Galaxy, Planetfall, A Mind Forever Voyaging and Trinity made text adventures a serious commercial genre. Infocom's parsers were the most sophisticated of their day. You could type "put the blue bottle on the second shelf", and the game would pick apart the sentence, work out the verb and the objects, check you actually had the bottle and could reach the shelf, and then (sometimes) understand you. The puzzle design that grew out of that capability was extraordinary. Brian Moriarty's Trinity, which sends you to the first atomic test, ends on a meditation about nuclear weapons that still lands as hard as anything in fiction.

The limitation was always the parser, however. It knew a finite vocabulary. Step outside it and you got "I don't know the word X." The promise of endless possibility kept crashing into a very finite implementation. As a result, players developed an almost telepathic bond with the parser, learning to phrase every action the way the designer expected. A game might understand "examine the chest" and stare blankly at "have a rummage through the chest." The genre turned into a vocabulary game wrapped around a story, and that ceiling, high as it was, was real. Do you remember that feeling of hunting for the one magic verb? By the end of the 1980s Infocom itself was gone, and in my view the commercial collapse owed as much to the limits of the form as to the rise of graphical games.

The Hyperlink Era (1993–2010)

The web finished off commercial text adventures, but it gave hobbyists a home. The parser didn't die so much as go underground. The annual Interactive Fiction Competition became a showcase, and some of the most celebrated work of the period came from there. Adam Cadre's Photopia won the 1998 competition and showed that a parser game could make people cry. Emily Short's Galatea (2000) built an entire game out of one conversation with a statue. Later, Inform 7 gave authors a way to write parser games in something close to plain English.

Meanwhile hypertext fiction took off. Tools like Twine (released by Chris Klimas in 2009) and publishers like Choice of Games let authors build branching narratives where clicking a link replaced typing a command. The player became a reader making decisions at key moments. Production costs fell to almost nothing, so a single author with no programming background could ship a complete interactive story.

Branching solved the parser problem by constraining choice entirely. There was nothing to misunderstand, because you could only pick from the options on offer. Even so, it created a new problem: every branch had to be written in advance. Stories became finite decision trees with an illusion of freedom. The most ambitious authors turned the constraint into an art form. Porpentine's Twine game howling dogs (2012), which Emily Short reviewed during that year's competition, used its narrow paths to make you feel trapped on purpose. But the constraint was always there. You never did anything the author hadn't anticipated.

The Hybrid Era (2010–2019)

Episodic adventure games from studios like Telltale and Quantic Dream, along with visual novels, brought branching stories to a mass audience. They dressed the branches up in 3D, with voice acting and cinematic camera work. The promise that your choices matter was sincere and partly kept. Underneath, though, the structure was still the branching tree. Plenty of players who loved these games could feel the edges of the form. Most choices made small cosmetic differences and only a handful made big ones. Sooner or later, everything funnelled back to the same trunk. "Clementine will remember that" became a meme for a reason.

Over in the tabletop world, online tools like Roll20 and Foundry made remote D&D workable. Live-streamed actual-play shows like Critical Role showed an enormous audience what real interactive narrative looks like: a human GM improvising in response to the players, with no pre-written branches and no false walls, producing stories nobody could have planned. That raised the bar for what interactive fiction's audience expected. Once you could watch actual-play sessions on YouTube, branching trees started to feel a bit dated.

The AI Era (2020–present)

Large language models changed what's possible. For the first time, a system could understand natural language and write a fitting narrative response, drawing on patterns learned from enormous amounts of human writing instead of looking things up in a table. AI Dungeon was the first widely played example. Nick Walton, then a student at Brigham Young University, built the first version at a hackathon in spring 2019, and when an improved version came out in December 2019, Aaron Reed's 50 Years of Text Games records that it drew 100,000 players within a week. It was rough, mind you. The state drifted, and characters could forget their own names halfway through a plot that was busy dissolving into nonsense. Yet it proved something nobody had shown before: a story engine with no pre-written branches and no fixed vocabulary could still produce coherent fiction in response to anything you typed.

The implications are huge. Look at the three obstacles that each defined an era of interactive fiction:

• The parser problem disappears, because you can say anything
• The branching problem disappears, because the story isn't written in advance
• The GM burden problem disappears, because an AI can run a session without a human doing hours of prep

Losing all three at once isn't an upgrade. It's a change of category.

EchoQuest is built on that foundation. Its Game Master runs on Claude, a far more capable model than anything AI Dungeon had in 2019, and it sits inside a structured game engine that holds the world together where the AI alone would drift. Your HP, inventory, quests and relationships live in real game state, and the dice are rolled by the system, not invented by the storyteller. Still, we're early, and I'd rather be upfront about the limits. Long-term memory is imperfect, and consistency across many sessions takes real engineering. Meanwhile, the emotional depth of a skilled human author at their best remains unmatched.

I've felt each of those limits personally. I once found that memory summaries had got stuck: once a game passed a certain length, every turn re-summarised the first ten turns and nothing after them. Games that weren't saved on the server simply forgot their oldest turns once the history filled up. Both are fixed now, and the same day I moved character progress onto the server so you can pick up from any device. I also rewrote the GM's instructions that day, because they read like a technical spec and the narration was coming back sounding like one. I talk about these limits openly because pretending they don't exist would insult the form's future.

What Comes Next

The path ahead points toward AI narrative that:

• Stays consistent across hundreds of sessions, with structured memory that doesn't fade
• Adapts its style and pacing to each player over time, learning which beats land for you in particular
• Generates original music and voices that keep step with the story in real time, and perhaps pictures one day
• Supports genuinely multiplayer adventures, where several players' choices collide through shared AI improvisation (that one's near the top of my own list)
• Reaches the emotional ceiling of human-authored fiction, not by mimicking it, but by being honest about what AI does well and sharpening those strengths

We're at the start of this arc, somewhere around the equivalent of 1978 for the parser era: a few interesting experiments shipped, with the medium's best work still ahead. The history of interactive fiction is full of moments when something unthinkable became ordinary within five years. I expect the next five to produce things that make our 2026 experiences feel as quaint as Zork feels now. Which era did you start in, by the way?

The best interactive fiction ever made hasn't been written yet. My bet is that an AI will write it, together with human creators, in response to the choices of a player who had no idea any of it was possible on the day they started.

Play the current state of the art →

Related reading: Browse all EchoQuest blog posts.