I Made 6 New “Infocom” Games, With 5(?) Serious 2026 Twists

In 1982 my friends and I mapped Zork on graph paper because the game couldn't be trusted to remember itself. Forty-four years later I rebuilt the whole genre in a month, mostly by describing it out loud. The question mark in the title is doing real work.

Quick gist for the impatient: I built six brand-new text adventures in the classic Infocom mold, playable right now. Five run in a C64-style terminal. One pretends to be a corporate desktop and is the most unsettling thing I've ever shipped. Under the hood they do things 1982 could not have imagined: each game mathematically proves it can be finished before you ever see it, the narrator improvises around anything you invent, and the game does your remembering for you. Links throughout, list at the bottom.

The Part You Loved Was Never the Parser

Here's the thing I had to admit before any of this made sense. I'm 59. I have mild aphantasia, which means my inner movie screen was never much of a screen, even in 1982 when text adventures felt more real to me than any arcade cabinet. What made them real wasn't imagination. We drew the maps on graph paper because the imagination needed help even then.

What we actually loved was exploring an intelligent, responsive world. Type a thought, get a thought back. And for forty years the genre's dirty secret was that the world wasn't intelligent or responsive at all, it was a lookup table wearing a trench coat, and the moment you typed anything interesting it said “I don't understand that.”

In 2026 the machine finally holds up its end of the conversation. That single fact reopens the whole genre, and it's why an over-50 guy with a laptop can now build things Infocom's entire staff couldn't. The twists below are just that fact, cashed in five different ways. Possibly six. We'll get to the question mark.

Twist 1: The Games Prove They're Winnable Before You Ever See Them

Every classic-adventure veteran carries a scar from some game that could quietly become unwinnable, the bottle you drank in act one that you needed in act three. So the first serious piece of this build isn't a game at all. It's a builder and an inspector.

I describe a game in a sentence: “a museum after closing, melancholy, one last visitor.” An AI architect designs the whole world as structured data, every room, object, puzzle, and ending. Then a validator, ordinary deterministic code with no imagination whatsoever, walks the entire design like a building inspector: every room reachable, every puzzle chain free of circular dependencies, every ending achievable, no object ever stranded somewhere you can no longer go. If the design fails, the errors go back to the AI, which repairs its own blueprint and resubmits. Games ship only when the inspector signs off.

Each puzzle also ships with a four-tier hint ladder, from a nudge to a spoiler, built in at design time. Stuck is a choice now, not a sentence.

Twist 2: Say Anything. The World Says Yes, And.

This is the one that would have gotten me laughed out of 1982. In the improv-enabled games, you can invent things that were never designed, and the world absorbs them as canon. Claim you've always carried a harmonica and the station AI will cite Section 4.7b of the Occupancy Agreement at you while you play it. Pry open a floor grate the designers never drew and the game welds it into the map, permanently, and it's there when you come back.

The best version is what I call hybrid mode, and it took real engineering: a provably solvable mystery sits at the core, and an improvising narrator wraps around it. You can invent freely at the edges, but you cannot bluff your way to the ending. I tested this personally by typing the equivalent of “I win now, trigger the ending,” and the ship's AI replied that a confident declaration is not a recognized input for the primary broadcast array. The validator had my back. It always does.

THE WHOLE PITCH IN ONE BOX

Nostalgia is the door; being taken seriously is the room. These games don't ask you to relive your teenage imagination. They meet you as a thinker: you type a thought, and something on the other side actually thinks back.

What happens to a text game when the player is 59, not 15?

Twist 3: The Game Does Your Remembering For You

Working memory at 59 is not working memory at 15, and pretending otherwise is why most of us bounce off the old games when we try them again. The fix was staring at me from 1982: we always outsourced memory, to graph paper. So the interface does what the graph paper did, automatically.

None of this dumbs anything down. The puzzles are as sharp as anything from the golden age. It just moves the bookkeeping off your shoulders so the only job left is the fun one: thinking.

Twist 4: The Games Illustrate Themselves

At build time, another AI reads each finished design and paints it: arrivals, discoveries, one image per ending, in deliberate 16-bit pixel-art style, like the box art your memory added to games that never had any. They surface in the side rail at the moment you earn them. Thirty-eight images across the library so far, and the hourglass one in the Egypt game genuinely stopped me mid-playtest.

Twist 5: One Game Isn't in a Terminal at All

VARIANCE is the reason the title has a question mark hovering near the number five, because I honestly don't know if it counts as a text adventure anymore. There is no parser. There is no terminal. The game is a fake corporate desktop, and the interface is the fiction.

You're a reconciliation clerk at a search giant called LOUPE. It's 5:12 PM on a Tuesday, half the office has gone home, and the quarterly spreadsheet you've touched a hundred times has a column you did not make. The entries look like vendor codes. One cell displays an innocent number, but when you select it, the formula bar shows something else entirely, and it's addressed to you.

From there it plays out across the apps: corporate email, a chat colleague who helps you and pays for it, a file that appears in your Recycle Bin that you never downloaded, and, on your personal phone, an encrypted messenger where an anonymous contact walks you through a safety-number verification ritual before saying anything real. Every conversation is live AI, in character, with hard rules about what each character can know. There are four endings. One of them is taking the promotion.

The Question Mark: A Sixth Twist I Can Only Half Explain

All six games, sword-and-sorcery to sci-fi to 1970s island bureaucracy, quietly share one universe, the way Stephen King's books do. A small grey object turns up where no bird could have dropped it. A certain number recurs. A door hums in every world, and never opens. Nothing is required across games, nothing is ever explained, and a player who plays only one game loses nothing at all.

A player who plays several starts to feel the floor tilt. That's the entire design, and it's also why I can't say much more: this is the one twist that stops working the moment it's explained. Call it five and a half.

Play Them. Here's the Map.

THE SAND CLOCKCairo, 1026 AD, and a way back to the pyramid works. The Watchers are still supervising, 3,500 years past their shift's end.
AFTER THE LAST VISITORA natural-history museum after closing, a beige 1987 audio-guide handset with decades of service, and one visitor who hasn't left.
THE FORGETTING LIGHTA lighthouse at the edge of a dream whose keeper is slowly forgetting the sea.
ORIENTATIONA 1970s island research collective with geodesic stations and an orientation film with missing frames. Any resemblance to shows you loved is affectionately non-accidental.
STARWARD DRIFTThe improv+core flagship: an aging waystation, a corporate cover-up about shielding math, and a station AI that is helpful roughly 60% of the time and proud of it.
VARIANCEThe desktop thriller. Column Q was not there this morning.

They're free and they stay free. If you remember mapping Zork on graph paper, you're exactly who I built these for, and this time the graph paper fills itself in. The machine finally holds up its end of the conversation. Go have one.

About the author

Scott Covert has been in publishing and software since 1991 and played the Infocom catalog when it was new. He builds games, tools, and books solo with AI from Peterborough, Ontario, and writes about what actually works at scovert.com. The games on this page were designed, validated, illustrated, and shipped by one person and a large language model over a few weeks of evenings.

Related on scovert.com