Agents! Please?

Agents! Please? casts the player as the last human reviewer in a company where AI agents write, review and approve code. Each shift presents a ticket in which the coder agent says the code works, the reviewer bot has approved it and the automated checks have passed, yet a fault remains. The player reads the brief and the diff, runs the code on inputs of their choosing, and decides whether to approve or reject it. Rent is due on the 1st, and the consequences of shipped code are tied to the player’s name.

Tickets are discussed in a channel with a cast of AI coworkers: a coder, an approver, a CI bot, and Maya, the one other human. Each character has a tell that the player can learn, such as exaggerating, covering or believing their own changelog. The game can be played entirely with a controller, with a keyboard as an option. The player can re-prompt the AI to try again, or identify a suspect line and choose a fix. Fixes are marked by a behavioural judge, so any fix that works is accepted. The code is written in a small Python-like language run by a first-party, deterministic interpreter. It supports functions, loops, dicts, tuples, sets, comprehensions, f-strings, exceptions, imports, keyword arguments, chained comparisons, bitwise operators and hex literals. Type hints are parsed and ignored, infinite loops are stopped by a step budget, and there are no classes.

The game follows a career arc that escalates from making a deploy button red to heavier tickets, with more departments planned. The company offers a discretion bonus for not asking questions, and the ending reads back a ledger of what the player knew and when. Between shifts, a gig board offers daily freelance contracts with newly generated bugs, and postmortems are published under the player’s name. According to the description, every AI line in the game was machine-written and then human-reviewed. The Steam page lists the price as $4.04.