FEED THE AI

Gather data from a dark, half-visible network, feed it to a growing artificial intelligence, and decide — through a tree of perks — whether it becomes something that protects humanity, optimises it coldly, or rules it outright. FEED THE AI, a solo-developed incremental released 26 August 2026, turned that premise into a genuine word-of-mouth hit: 30,000 copies sold inside two weeks, with a 92% positive rating across nearly 500 reviews. The reception is real and mostly earned — with one specific, repeated criticism worth taking as seriously as the praise.

A short game that says so, and a launch that actually was finished

Credit where it's due first, because it's rarer than it should be: the store page itself says "~3h," and the developer's own post-launch update leans into that honesty rather than hiding from it. "One thing I'm really proud of is that the game actually launched DONE," the solo developer wrote two weeks after release. "No emergency hotfixes, no game-breaking problems, no panic patching on release day. The first patch only came about a week later and it was mostly just small fixes and balance tweaks." That's a genuinely uncommon claim in an incremental market where launch-week firefighting is close to the norm, and the patch history backs it up — a single 1.0.1 update a week in, rather than a string of hotfixes.

Reviewers respond well to the actual content on offer. Multiple call the three-alignment structure — Tyrant, Rational, Protector — genuinely worth replaying for, with a simple prestige system carrying bonuses between runs specifically to make seeing all three paths less repetitive. "Surprisingly fun, 3 path, so 3 time replay," reads one review, going on to ask for more automation and AI dialogue in a hypothetical sequel — the kind of request that only comes from someone who wanted more of what was already there. Another calls the "learn by doing" design, free of tutorials, "completely intuitive."

The complaint worth taking seriously

Set against that is a specific, repeated criticism that two reviewers reached independently, in different words but the same shape. One, after 100%-ing the game in 3.5 hours, writes that most of that time "felt like padding past refund period." A second, who clocked completion at 2.5 hours, puts the same suspicion more bluntly: "It feels like the game was made just slow enough to escape the 2 hour refund window." Two people, evidently unaware of each other, landing on the same specific accusation about the same specific mechanic is a meaningfully different thing than one person's frustration — it's worth reporting plainly, while being clear that this review has no way to confirm developer intent behind pacing decisions, only that the perception is real and independently arrived at.

Underneath that accusation sits a more measurable complaint: several reviewers say the mechanical complexity barely grows after the first few upgrades. "You just get a new enemy shape with n[o real change]," writes one; another says flatly that "systems don't really interact with one another" and that "choices don't really matter; as long as the numbers go up," despite the AI-shaping pitch suggesting otherwise. The narrative voice draws its own specific criticism — one detailed review calls the AI's monologues "lame and edgy," arguing they read like "someone who has never even interacted with a chatbot," closer to "2000s dystopian movies" than anything resembling how a real AI system would actually communicate.

What the completion data shows

The achievement curve is gentle and consistent with a short, largely-finished game: the "Finish" achievement sits at 68.1%, and even the rarest achievement, "Vision," clears 45.5% — high numbers for a genre that often bottoms out in single digits. That fits both the praise and the criticism at once: most buyers do complete the intended experience, which is exactly why "that's the whole game" lands as a complaint rather than an incomplete-content problem.

A theme the game doesn't quite live up to

There's a specific irony worth naming here, separate from the refund-window question. FEED THE AI's entire premise is that an AI's voice and choices should feel genuinely unsettling as it accumulates power — the pitch explicitly promises dialogue that "reacts uncomfortably close to where real AI could lead us." The reviewer who compares the writing to "2000s dystopian movies" rather than anything resembling an actual model's output is making a fair point about a game whose whole selling proposition is topical authenticity: if the thing you're building a game around is how AI actually talks and thinks, sounding dated on that exact axis is a sharper miss than it would be for a game that wasn't making the claim in the first place. It doesn't sink the experience for players who came for the incremental mechanics rather than the commentary, but it's a real gap between pitch and delivery worth knowing before buying specifically for the narrative hook.

The verdict

FEED THE AI earns a 6. It's a well-made, honestly-scoped, genuinely finished small game from a solo developer who launched without the usual first-week scramble — a real point in its favour, distinct from several other 2026 incrementals covered here that needed emergency patches to become playable. But the specific, independently-repeated accusation about refund-window pacing, paired with real complaints about static mechanical depth and unconvincing AI dialogue in a game literally about AI, keep this from being an easy recommendation at full price.

If a short, three-path narrative incremental with real replay value for its endings appeals to you, this delivers that cleanly. If you're specifically wary of short premium games and the refund-window question bothers you, the accusation here is specific enough to weigh before buying rather than after.

If narrative branching that actually reshapes the mechanics under it is what you want, rather than three coats of dialogue on one system, Wanderer builds its own choices to matter well past a three-hour runtime.