Early Access and the Player as QA

Steam Early Access opened in March 2013, and the sales pitch has stayed remarkably stable for over a decade: pay now, play the unfinished thing, help shape it. The common shorthand is that the players are doing QA. That shorthand is wrong in a way that’s worth taking apart, because the thing Early Access actually does is more useful than QA and more expensive than anyone admits.

QA is a job, and this isn’t it

Quality assurance is an adversarial discipline. A tester is paid to hate your game for eight hours a day. They work from a test plan, they try the input you’d never try, they hold the stick in a direction no reasonable person would hold it, and when something breaks they write a reproduction case precise enough that a programmer can follow it. The skill is in the repro. Anyone can find a crash by accident; a tester finds it on purpose and then finds it again.

Early Access players are the opposite population on every axis. They are self-selected, they have paid money, and they are rooting for the game. They are enthusiasts, which means they play the way an enthusiast plays — down the intended line, with goodwill, forgiving the rough edges because roughness was the deal. They do not write repro steps. They write “it crashed lol”. They are also, crucially, not distributed across the hardware space in any useful way; the driver bug on the unfashionable GPU stays unfound because the people who own that GPU aren’t buying half-finished games at launch.

So Early Access is bad at the thing it’s named for. What it’s extremely good at is something else entirely: playtesting at a scale no studio could ever afford. Ten thousand people playing your economy for a hundred hours each will find the degenerate strategy. They’ll find the weapon that trivialises the third act, the resource that’s worthless, the difficulty spike at hour six that makes a third of them stop. That’s design data, and it’s genuinely irreplaceable — a publisher’s focus group of twelve people in a room for an afternoon cannot produce it, and neither can the team, because the team has lost the ability to be surprised by their own game.

The distinction matters because it predicts which games Early Access helps. It helps a game whose engineering is sound and whose balance is open. It does nothing for a game that’s fundamentally broken, because the audience arrives with the wrong instincts to find out how.

Novelty is non-renewable

Here’s the cost nobody itemises. You can play a game for the first time exactly once. The first hour of a good game — the confusion, the dawning sense of how the thing works, the moment the loop closes and you understand what you’re doing — is the single most valuable experience the game will ever produce, and it’s strictly non-renewable. Early Access spends it on a build the developer intends to throw away.

Baldur’s Gate 3 is the clean illustration. Larian put Act 1 into Early Access in October 2020 and shipped 1.0 in August 2023. Three years is a long time to know a first act by heart. The people who played it most — who were most useful to Larian, who generated the most data — walked into the finished game having burned the opening on a version that got rewritten underneath them. They paid for the privilege of not being able to meet the game properly. That’s a real transfer of value, and it flows from the audience to the studio, which makes the discount on the Early Access price look less like a favour and more like a wage.

None of which makes it a bad deal. It makes it a deal, with two sides, and the industry’s habit of describing it as a gift to the community obscures the trade.

The thesis test

Compare two Early Access runs. Hades entered Early Access on the Epic Games Store in December 2018 and released in September 2020. On day one of Early Access the loop was already there: you fight your way out, you die, you talk to people, you go again, and the conversation system that made the game famous was already load-bearing. Supergiant used two years to add rooms, weapons, bosses and dialogue inside a frame that was already built. Every update tuned a thing whose purpose was settled.

Double Fine’s Spacebase DF-9 went into Early Access in October 2013 and shipped a 1.0 in October 2014 that most of its audience read as an abandonment — the scope had been cut back to something the money could finish. Talent doesn’t explain the gap — Double Fine has plenty. What explains it is that Spacebase was still discovering what it wanted to be while an audience watched, and design discovery is a process that looks exactly like failure from the outside. Every promising direction abandoned is, to a paying customer, a broken promise. Early Access punishes the studio that hasn’t decided yet, which is precisely the studio most likely to need the money.

Dead Cells passes the same test that Hades does. Motion Twin went into Early Access in May 2017 with the combat already excellent and the structure legible, spent fifteen months adding to it, and released in August 2018. The rule generalises: Early Access works when you already know what the game is and need help finding out how it should feel. It fails when you need help finding out what it is.

The permission to ship broken

The interesting ancestry runs back further than 2013. Look at what the model requires: the ability to patch. Without patching there is no Early Access, no roadmap, no 1.0 as a distinct event from release.

I loaded games off cassette for most of the eighties, and the thing about a tape — or a cartridge, or a floppy in a box in a shop — is that the ship date is the last date. What went in the box was the game forever. That constraint was miserable in obvious ways and disciplined in one non-obvious way: a bug that shipped was a bug that stayed, so the bug had a cost the accountants could see. The C64 canon holds up partly because those games had to be finished, in a sense the word has quietly lost.

Patching removed the cost, and the incentive followed the cost. This isn’t a moral complaint about modern developers being lazy — the same patching pipe that permits a broken launch also permits Hades’ two years of improvement, and I’d take that trade every time. It’s a structural observation: once fixing it later became possible, shipping it broken became survivable, and anything survivable eventually becomes standard. The shareware model that carried Doom sits halfway between the two eras — id gave away a finished episode and sold the rest, which is a promise about a thing that already existed rather than a promise about a thing that might.

Valve, to its credit, says the quiet part in its own documentation: buy an Early Access game based on its current state, because some of them never finish. That’s an honest warning and also an admission that the roadmap is unenforceable. Nothing binds a studio to the plan. 7 Days to Die entered Early Access in December 2013 and reached 1.0 in July 2024 — over a decade in which the word “unfinished” did an enormous amount of work. DayZ took five years, from December 2013 to December 2018. In both cases the game people bought and the game that eventually shipped are related the way a demolition site is related to a house.

What honest looks like

The model isn’t going anywhere, and it shouldn’t. It has funded a decade of games no publisher would have greenlit, and the alternative — a world where every ambitious mid-size game needs a suit in Guildford to approve it — was worse. What I want from it is smaller than reform.

Price the state of the build. If the game is a third finished, charge a third. Say what you don’t know. Treat the roadmap as a statement of intent with the confidence intervals visible, because the audience can handle “we’re not sure this will work” far better than they can handle discovering it in a patch note eighteen months later. And when the design question is settled, ship — the long Early Access tail is usually a studio that can’t tell the difference between improving a game and being unable to let go of it.

The best argument against treating Early Access as inevitable is that Balatro sold in the millions after releasing, in February 2024, as a finished game. One developer, no roadmap, no eighteen-month tuning period with a paying audience in the room. It arrived knowing exactly what it was. That’s still allowed, and it’s still the strongest position a game can launch from — the tape-era discipline, running on hardware that would let you patch and simply not needing to.