The Cost of AI-Generated Code: Solving Developer Decision Fatigue

Coding agents are giving everyone decision fatigue

Pratima Arora, CPTO of Smartsheet, highlights a critical shift in the software development lifecycle. Research shows that 80% of AI-generated content requires manual editing before finalization.

Why This Matters

While AI agents have reduced the cost of writing code, they have increased the cognitive load on the review and verification stages of the SDLC. The technical reality is that high-velocity code generation creates a ‘density of work’ where developers spend more time gathering context and making high-stakes judgement calls than actually coding. This imbalance leads to decision fatigue, increasing the risk of human error—such as source code leaks—when reviewers become sloppy due to burnout.

Key Insights

• Automation intensity for enterprise users grew 55% year-over-year (Smartsheet, 2026).
• The ‘Builder’ concept enables non-engineers to prototype quickly using tools like Claude and Cursor to solve customer problems.
• Goodhart’s Law is resurfacing as organizations mistakenly track productivity via token usage or percentage of AI-written code rather than outcomes.

Practical Applications

• )Use case: Smartsheet designers use Claude and Cursor to build front-end prototypes based on design systems before engineering handoff.
• Pitfall: Measuring productivity by lines of code produced by AI agents, which can lead to a bottleneck where one ‘superstar’ produces 7X more code than teammates, overwhelming the peer review process.

References:

• https://stackoverflow.blog/2026/05/21/coding-agents-are-giving-everyone-decision-fatigue/