Chaos reports up to 10x output gains from AI agent adoption

Christian Radowski, Deputy CTO at Chaos, reports that AI coding agents have been rolled out across the company's approximately 250-person engineering organization. While the company does not aim for a fully automated dark factory due to the complexity of its software, Radowski observes that developers who fully adopt the tools produce output that is 4x, 5x, or even 10x higher than those who do not. This adoption has reduced development timelines for certain products from 12 to 18 months down to 2 to 3 months. However, Radowski notes that the cognitive load on engineers has increased rather than decreased, as they must now review significantly more code and adapt to tooling changes every four to six weeks. The company mitigates quality risks by maintaining strong QA processes and injecting specific context, such as design decisions and terminology, into agent prompts to ensure code maintainability. Radowski argues that inactivity is the only failure state, as token costs are predictable and the rapid pace of model improvement makes waiting for a clear winner unviable. The company's AI transformation was initiated by a combination of top-down C-level strategy and bottom-up experimentation by AI-native engineers. The CTO, Vladimir Kojetsov, focuses on research and bleeding-edge technology, while Radowski manages daily business and engineering delivery. Radowski states that leadership in this context requires accepting a high degree of uncertainty, as the technology landscape changes rapidly and no single expert has all the answers. He describes his role as setting a clear goal image and allowing teams to experiment, then adjusting course based on feedback. He identifies key warning signs of organizational overload, such as a shift in discussion culture from questions to blockade attitudes, or when initiatives are consistently delayed. Radowski emphasizes that leaders must communicate that changes are driven by necessity and mutual benefit, not just KPIs, to maintain team morale and acceptance. Radowski assesses the impact of AI on the company's business model, which serves the architecture, engineering, and construction sectors as well as VFX. He notes that while AI tools have made basic visualization a table-stake feature, the AEC industry is conservative and slow to change due to the high cost and long horizon of decisions. Consequently, Chaos is not heavily impacted by the SaaS apocalypse or customers building their own tools via vibe-coding, as the industry values trust and established customer relationships over cost savings from free clones. Radowski acknowledges that reverse-engineering tools exist for CAD systems but argues that the complexity of enterprise software makes 1:1 replication unlikely and that the lack of maintenance and security guarantees for free clones is a significant barrier. He believes that domain knowledge and the ability to understand industry workflows remain the primary moats for companies like Chaos.