The New Engineering Reality: Why AI-Pilled Teams Are Shipping 8x Faster

Fiona Fung, who leads the teams behind Claude Code and Cowork at Anthropic, describes a fundamental change in software engineering. Coding is no longer the primary bottleneck. Instead, the new challenge is managing complexity and maintaining quality at high speed. Fung explains that successful teams are moving away from traditional six-month roadmaps toward just-in-time planning and automated, asynchronous agent routines. This shift carries a hidden risk of loneliness and skill loss, as the deep focus of manual coding gives way to the high-level orchestration of AI agents. For engineering leaders, the advantage lies in shifting focus from token usage to outcome-based verification. Those who master this transition will build a durable competitive edge, while those who resist will likely face operational decline.

The Shift from Manual Coding to Orchestration

Engineering has evolved from a solitary, human-centric craft into a high-agency, AI-assisted discipline. At Anthropic, engineers are shipping eight times as much code per quarter as they were in 2025. This is not just a marginal gain; it represents a massive increase in the output of a small team.

The important dynamic here is that as the cost of generating code drops to near zero, the value of the builder shifts from writing lines to verifying outcomes. Fung notes that the most effective engineers are now those who possess deep systems expertise to verify what the AI produces, combined with the product sense to understand if the output actually solves a user problem.

"Coding is no longer the bottleneck; it has lifted the ceiling of what anyone is able to do. Everything is now possible in theory; now it is about how ambitious you can be."

-- Fiona Fung

The Bad vs. Sad Framework for Quality

When teams ship 8x faster, traditional code review becomes a major bottleneck. Fung’s team at Anthropic has responded by evolving their quality framework. Instead of relying on manual oversight, they focus on proactive quality detection by categorizing issues into "bad" (unrecoverable errors, crashes) and "sad" (recoverable pain points, UX friction).

By empowering teams to define these thresholds themselves, Anthropic maintains high agency while ensuring that automated agents have a clear framework for what good looks like. This is an evolution of Test-Driven Development where the model is tasked with failing a test before writing the code, but scaled to the entire product surface.

"The more that we can automate to check in the framework for what good looks like, Claude is very good when you give it a framework to validate against those frameworks."

-- Fiona Fung

The Hidden Cost of Asynchronous Agents

Fung highlights a critical downstream effect of this new way of working: the rise of the context-switching tax. As managers and engineers deploy fleets of agents to handle feedback channels, bug fixes, and planning, they face an endless stream of asynchronous inputs.

The system forces a trade-off. While agents handle the doing, the human role has shifted to reviewing. This creates a hidden cost: the loss of the deep, immersive flow state that previously defined the engineering experience. To combat this, Fung’s team has implemented pairwise programming lunches and hackathons, not just for code, but to reconnect the human element of the team. The payoff for this effort is a culture that avoids the dog in the burning room scenario, where managers are too busy to notice that their team processes have become unsustainable.

Key Action Items

• Implement Just-in-Time Planning: Move from six-month roadmaps to monthly, lightweight planning cycles. Review these weekly to ensure they still serve the current reality. (Effective immediately)
• Adopt the Bad vs. Sad Framework: Define clear, team-specific thresholds for what constitutes a critical failure versus a recoverable pain point to prioritize automated quality checks. (Over the next quarter)
• Audit Your Processes for Obsolescence: Explicitly give your team permission to kill any process that no longer serves its purpose. If it feels like a tax, it likely is. (Ongoing)
• Mandate Dogfooding for Leaders: If you are a manager, ensure you are using your own product daily. Use the product to find bugs and UX gaps that metrics and dashboards miss. (Immediate investment, pays off in product quality)
• Build Onboarding Buddies for Managers: New managers should spend their first few weeks as individual contributors to learn the current codebase and tools before taking on full management responsibilities. (12-18 month investment in team rapport)
• Create Human Connection Points: To combat the loneliness of agent-assisted work, schedule regular pairwise programming or collaborative sessions to ensure knowledge sharing and team cohesion. (Over the next quarter)