AI can generate code faster than ever, but getting it tested and reviewed and into production is a different challenge. Engineering teams are increasingly using AI agents to write code, fix bugs, and build features. But faster coding does not necessarily mean faster software delivery. As code generation accelerates, reviews pile up, testing slows down, and teams struggle to coordinate the growing volume of changes. At TechSparks 2026, Tushar Mathur, CEO of Forge Code, explored this problem in his session, Anatomy of a software factory. He explained how AI-native engineering organisations can move beyond automating individual coding tasks to building systems that manage the entire software development lifecycle. “Just because code is becoming easy and faster to generate doesn't mean your overall SDLC becomes faster,” Mathur said. From coding assistants to software factories The traditional software development process focused heavily on writing code efficiently. Teams spent time defining requirements, preparing specifications and designing architecture to reduce the effort needed during implementation, he said. With AI taking over more coding work, the bottlenecks are moving elsewhere. Mathur described how organisations can end up with large volumes of pull requests awaiting review, longer testing cycles and more bugs than existing regression suites can catch. Hiring more engineers or running more AI agents is not necessarily the answer. More people can increase coordination costs, while adding agents without effective feedback loops can generate code that overwhelms the systems responsible for reviewing and testing it. “The reason why this happens is because there's no signal,” Mathur said, explaining why simply connecting automated steps does not create an effective software factory. An assembly line that learns Mathur presented Forge Code as a software factory that coordinates AI agents across the development lifecycle. The process begins with triaging requests, creating specifications and implementing features. Agents then verify changes before they move towards deployment, with human engineers stepping in when decisions or feedback are needed. The system also monitors production, using the resulting insights to inform future tasks. This creates a feedback loop in which the factory can identify problems, improve its workflows and generate new work from what it learns. Mathur said Forge Code was running around 100 agents in the background during his presentation, working on features, compatibility and bug detection. He outlined six principles for an effective software factory: autonomous operation, scalability, cost efficiency, measurability, self-improvement and alignment with business goals. The last principle is particularly important. Instead of treating every request as a coding task, the factory should work towards outcomes such as improving customer retention or meeting compliance requirements. Teams can then prioritise features based on the value they deliver. Engineers become supervisors In this model, engineers spend less time directing every task and more time monitoring outcomes, reviewing signals and improving the system. Business leaders can also contribute requests and priorities directly, allowing the factory to translate organisational goals into a development backlog. Writing code faster is great. Shipping better software is the real win. For engineering teams, the trick is turning AI-powered speed into reliable products, measurable gains and business results.