AI Product Velocity, Product Taste, and the End of Code Scarcity
Anthropic's Head of Product Kat Wu reveals how development cycles have compressed from quarters to days, rendering traditional roadmaps obsolete. Insights cover the rise of 'product taste' as the scarce skill, the 'model eats harness' dynamic, and the necessity of 100% reliable automation for genuine leverage.
The Compression of AI Development: Insights from Anthropic's Kat Wu
The AI product landscape is undergoing a radical compression of development cycles. Kat Wu, Head of Product at Anthropic, outlines how shipping timelines have collapsed from quarters to days, necessitating a fundamental shift in how product managers and engineering teams operate. This analysis explores the transition from code-heavy execution to judgment-driven strategy, the critical role of 'product taste' in an era of cheap code, and the operational mechanisms that enable Anthropic's relentless velocity.
Velocity Over Roadmaps
Product development at the frontier of AI has abandoned multi-quarter planning in favor of weekly or daily iteration. Timelines for features have shrunk from six months to as little as one day. This acceleration requires 'low process' frameworks that remove barriers to shipping. Mechanisms like 'Research Preview' branding allow teams to launch early, gather feedback, and pivot without the overhead of long-term commitments. Cross-functional friction must be eliminated; engineering, marketing, and documentation must operate in tight, automated loops to support rapid releases.
Product Taste: The New Moat
As the cost of writing code approaches zero, the ability to decide what to build becomes the primary competitive advantage. 'Product taste'—the judgment to prioritize features, define user value, and anticipate model capabilities—is now a rarer and more valuable skill than technical implementation. Hiring strategies are shifting toward engineers with strong product intuition, blurring the lines between roles. The human value proposition lies in common sense, stakeholder management, and the ability to navigate ambiguity, not in generating syntax.
Mission-Driven Execution
Anthropic's success is attributed partly to a unifying mission that simplifies decision-making. A clear, non-negotiable goal (e.g., Safe AGI) provides an instant tie-breaker for resource allocation, allowing the organization to move with unified speed. Teams are empowered to sacrifice individual product KRs for the broader organizational mission, preventing scope creep and maintaining focus. This mission-centric approach enables rapid, cross-functional execution that outmaneuvers competitors distracted by broader, less aligned portfolios.
The Automation Threshold
AI automation delivers value only at near-perfect reliability. Partial automations that achieve 95% accuracy often introduce more friction than they save, requiring manual oversight that negates time savings. The strategic imperative is to invest the necessary effort to refine workflows until they reach 100% reliability. Only then can teams unlock true leverage, offloading repetitive tasks to AI and focusing human capital on high-impact, creative problems.
Strategic Imperatives for Leaders
Organizations must adapt to the reality that models will eventually 'eat the harness.' Features built as crutches for current model limitations must be designed for obsolescence, with the ability to strip away complexity as intelligence improves. Leaders should encourage a culture of 'just doing things,' where employees are empowered to act across boundaries to solve problems. Finally, building and using AI tools in daily workflows is essential; prototypes do not generate leverage, but integrated, daily-use apps drive real productivity and feedback loops.
Key insights
-
Product development cycles have compressed drastically, with shipping timelines dropping from six months to weeks, days, or even hours. Success now depends on 'low process' environments that empower teams to ship rapidly, utilizing strategies like 'Research Preview' branding to reduce commitment friction and enable continuous iteration.
Impact: Startups must abandon rigid roadmaps in favor of rapid experimentation loops to remain competitive; organizations that fail to reduce shipping friction will fall behind in the AI economy.
-
As coding costs plummet, 'product taste'—the ability to decide what to build and prioritize effectively—becomes the scarce, high-value asset. Hiring engineers with strong product intuition is more efficient than maintaining siloed PM roles, as the core challenge shifts from implementation to judgment and prioritization.
Impact: Venture capital and hiring strategies should prioritize cognitive judgment, domain expertise, and decision-making capabilities over pure technical skills, reshaping the value proposition of workforce roles.
-
Product features often act as crutches for current model limitations; as models improve, these harnesses become obsolete. The 'model eats your harness' dynamic means products must be designed to strip away complexity over time, evolving toward simpler interfaces as underlying intelligence increases.
Impact: Product roadmaps must be provisional; investing heavily in complex harnesses that models will soon natively handle represents a misallocation of resources and technical debt.
-
A unifying mission enables rapid, unified decision-making by providing a clear framework for trade-offs. Teams willing to sacrifice individual product KRs for organizational goals can execute with higher velocity and focus, avoiding the distraction of fragmented priorities.
Impact: Companies with clear, mission-driven structures can outmaneuver competitors by reducing internal debate, aligning cross-functional efforts instantly, and maintaining strategic discipline.
-
AI automation is only valuable at 100% reliability. Partial automations create more friction than they save, requiring manual oversight that negates time savings. The strategic focus must be on refining workflows until they are fully autonomous to unlock genuine leverage.
Impact: Businesses should delay deploying AI workflows until they achieve near-perfect accuracy, ensuring positive ROI and preventing user fatigue associated with unreliable tools.
-
Evaluation frameworks are an underappreciated product tool that effectively defines success metrics. A small set of high-quality evals quantifies progress and guides model alignment more efficiently than extensive documentation, serving as a core artifact for product definition.
Impact: Treating evals as primary product artifacts streamlines development, aligns engineering and product teams on objective metrics, and reduces ambiguity in AI-driven feature delivery.
Action items
-
Implement a 'Research Preview' launch strategy for new features. Ship early with clear branding that indicates experimental status, allowing for rapid feedback collection and pivoting without locking in long-term commitments or support overhead.
Impact: Accelerates time-to-market, reduces risk, and enables data-driven iteration in an environment where model capabilities shift weekly.
-
Audit workflows to identify repetitive, low-creative tasks and deploy AI agents to handle them. Commit to iterating on these automations until they reach 100% reliability, ensuring they truly save time rather than requiring manual correction.
Impact: Unlocks significant bandwidth for high-value strategic work and increases individual productivity leverage by eliminating friction caused by partial automation.
-
Establish tight integration loops between engineering, marketing, and documentation. Empower engineers to trigger automated support from partner teams immediately upon feature readiness to eliminate cross-functional bottlenecks.
Impact: Reduces shipping latency by removing dependency waits, enabling the organization to capitalize on rapid model improvements and market opportunities.
-
Build and integrate AI tools into daily operational workflows rather than treating them as prototypes. Real usage generates the feedback necessary to refine products and realize tangible productivity gains for the business.
Impact: Drives deeper product-market fit and ensures AI adoption translates to measurable business outcomes rather than remaining a novelty.
-
Develop a focused set of high-quality evaluations for new features to define success criteria. Use these evals to track progress and identify gaps between model performance and product goals, treating them as core product artifacts.
Impact: Aligns teams on objective metrics, reduces ambiguity in AI development, and provides a quantifiable basis for prioritizing model improvements and feature adjustments.
Quotes
“As code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write.”
“The timelines for a lot of our product features have gone down from six months to one month and sometimes to one week or even one day.”
“If an automation doesn't work 100% of the time, it's not really an automation.”