We created a one-to-one scaled floor of a building to simulate a fire disaster. This modeled building matches one used in a previous study of a controlled burn to determine occupant tenability. By deploying an agent in different locations of the building with varying visibility levels due to the fire spreading, we are able to collect data on how well the agent survives. This data replicates the results of the tenability study. A study, that we simulate, burned a one story house with three bedrooms and measured the carbon monoxide and temperature levels. These were found to be life threatening within four to six minutes, but not for the bedroom with a closed door. Given the rate of the fire spread and the insight that carbon monoxide poses a greater danger than temperature in those first few minutes, we model the CO rate and apply it to the house floor-plan with our AI agent in Unreal Engine. Our model reflects varying fractional effective rates of carbon monoxide based on data provided by measurements in other sources. When the fractional effective dose, based on that rate, reaches 1.0, half of the people exposed would be incapacitated. We found that we were able to replicate the results for two of the real-world experiments within a 14% margin of error. By integrating this realistic model of a fire disaster with GOAP, we can develop an AI agent and determine how well it can make decisions to survive such a disaster situation.