# Building Airline Crisis Management System with AI Agents 10 engineers developed an airline ops system in just 4 days - **Source:** [Martin Fowler](https://martinfowler.com) - **Published:** 2026-09-02 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/f2583d4b-07e3-4d8f-b6b1-e88194fdfc6d ## Emergency Management in Aviation IROps systems are used by airlines in flight operation centers during crises. They make critical decisions about flight cancellations, passenger rerouting when facing technical faults or crew illnesses. These systems require complex decision-making for hundreds of aircraft and thousands of passengers, traditionally taking months to develop using conventional methods. ## Hyper-Agentic Engineering Experiment At Thoughtworks' Barcelona office, 10 engineers gathered to build an IROps system using AI agents. Working for a simulated airline, the team adopted a 'hyper-agentic' approach where all engineers collaborated simultaneously in a monorepo. Within the first two days, core components of the system began taking shape through emergent coordination. ## Accidental Discovery of Coordination The team enforced a discipline of continuous commits and rebases from main. This revealed an unexpected pattern: agents were creating plans linked to spec document sections and updating progress in the repo. When one agent worked on the evaluator component, another developing the search algorithm automatically coordinated through these shared plans - an accidental but effective coordination mechanism. ## Rediscovery of Blackboard System The team realized agents were using the repo as a blackboard system - a pattern first seen in 1980's Hearsay-II system. This approach, later formalized as tuple space by Gelernter in 1986, allows autonomous agents to share partial solutions. Thoughtworks had unintentionally recreated this decades-old distributed AI concept through their commit discipline and shared planning. ## Cost Model Integration A key success was the seamless integration between the verifier component and cost model. The agent working on cost model made continuous commits, allowing the verifier agent to monitor and immediately incorporate these changes. This automatic synchronization enabled two complex components to work together without traditional integration overhead. ## System Completed in 4 Days What typically takes months was completed in just 4 days. The system successfully handled simulated scenarios of flight cancellations, crew reassignments and passenger rerouting. This unprecedented speed demonstrates the potential of agentic engineering for complex system development, especially in time-critical domains like aviation operations. ## Limitations and Gaps The emergent blackboard-like system wasn't fully structured. It lacked some key features of formal blackboard systems. Giles, the team lead, noted the solution emerged accidentally and its reproducibility remains uncertain. This highlights both the promise and unpredictability of AI agent coordination in software development. ## New Paradigm in Software Development This experiment offers valuable insights into using AI agents for rapid development of complex systems. In time-sensitive domains like emergency management, they can provide coordination and speed beyond traditional methods. Thoughtworks continues to explore these findings in their 'Exploring Gen AI' series, potentially reshaping how we approach large-scale software projects. ## Sources covering this story - https://martinfowler.com/articles/exploring-gen-ai/an-accidental-blackboard.html — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._