Autonomous agents enter your online game, play the live build, and file what broke — with the repro a manual bench rarely has time to write.
Played across
Playtest floor
A run is a cloud session. The agent loads your build, joins the mode you care about, and writes down what it could and could not finish. Los Angeles built, for studios who cannot staff every night.
Capabilities
Play
You name the loop — place a match, spend the currency, finish the raid. The agent finds a way through, and a way it fails. It is not replaying last Tuesday’s click path.
Record
Input trail, build id, and a short capture stay attached to the finding. A producer can see the moment without replaying the whole night to catch it once.
Memory
Failures from the last run are first in line. New content is still explored. The floor does not start from a blank checklist every time the branch lands.
Grant applied twice in the night market. Session stopped at checkout.
Same stall is first. Then the agent walks the new seasonal stall.
Solutions
Same agents, different nights. Pick the lane that matches how your game actually ships.
Live service
Events, shops, and balance passes get a run while the patch is still internal. The finding lands with the stall, the price, and the account state.
Launch week
The modes you cannot staff overnight still get a session: fresh account, returning account, and a party that fails to fill.
Between merges
A quiet run after the branch lands. Regressions from the previous floor are replayed before anyone opens the build on Monday.
The agent bought the night-market bundle twice. The wallet debited once. Clip attached, 00:14, from the second confirm.
Critical · economyScripted QA repeats a path. These agents keep a working picture of the session — objective, inventory, other players — and notice when the build stops matching that picture.
The finding is the mismatch, plus how the agent got there. That is why a run can cost less than staffing the same night by hand, and why the ticket is usable the next morning.
Next build
Send the game, the mode, and the hour you want it played. contact@getplaytest.ai
Playtest AI is a playtest floor for online games. You point it at a build and a mode. Agents join, play toward an objective, and come back with what finished, what broke, and what they never reached.
A demo run starts with one mode you already know is expensive to cover by hand.
The same floor covers three nights that usually fall to whoever is still online.
People still own taste, feel, and the calls a clip cannot make. The floor is there so those people are not also the only ones who played the build.
An agent keeps a small picture of the session: the objective, what it holds, and what other players just did. When the build stops matching that picture, the mismatch becomes the finding.
Scripted paths still have a place. This floor is for the nights when the script no longer describes the game.
A demo is one mode of your online game, played by agents, with the findings read together. No slide tour standing in for a session.
Playtest AI is based in Los Angeles. Write with the game, how often you ship, and whether the pain is live ops, launch, or the nights between merges.
We read every note. If the game is a fit, the next step is a single mode on a real build — not a long qualification call.
