AI Writes Code Fast but Cannot Own Production Systems, Engineers Warn

·1 views

AI coding tools like Claude Code and Lovable can generate working prototypes in minutes, but engineers warn they fall short once software hits real-world production environments. IBM researcher Andy Anderson spent four months building a Kubernetes dashboard with AI assistance, only to lose more time reverting broken builds and fixing cascading errors than he would have spent coding manually. A large-scale study of over 302,000 AI-authored commits across nearly 6,300 GitHub repositories found that more than 15% introduced at least one issue, with 22.7% of those problems persisting long-term as silent technical debt. AI tools lack the contextual understanding needed to anticipate traffic spikes, dependency failures, or system-specific constraints that define production readiness. Engineers conclude that while AI accelerates prototyping, human ownership and judgment remain essential for building systems that survive reality.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Teen founders skip graduation to join SF accelerator, building AI efficiency startup

James Yang and Anish Paleja, both 18-year-old Canadians from Ontario, skipped their high school graduation to attend Founders Inc.'s Off Season II accelerator in San Francisco. The two met at a Major League Hacking hackathon at age 15, when neither could code nor had any product development experience. They have since founded Brevitas, a startup focused on reducing token waste in multi-agent AI systems to help companies cut costs on AI workflows. Despite the program being only six days old, the pair say they have already achieved roughly six months' worth of progress. Both are prolific hackathon participants, with Paleja having attended 36 and Yang 14 in the past year alone.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Systematically Read and Audit a Solidity Smart Contract You've Never Seen

A developer has outlined a structured method for reading unfamiliar Solidity smart contracts, arguing that starting from line one wastes attention on boilerplate before reaching critical logic. The approach prioritizes scanning for asset-movement functions first, then mapping access controls to identify who can trigger those functions and under what conditions. From there, the method involves reviewing state variables for unexpected write paths, checking whether external calls follow the checks-effects-interactions pattern to catch reentrancy risks, and finally auditing arithmetic for precision or decimal errors. The author also recommends using AI models to accelerate specific steps — such as generating access-control tables — rather than prompting them to find all bugs at once. The goal is to reduce the time needed to understand a contract's risk surface from a full day to roughly an hour.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Google DeepMind VP Says He Has Stopped Writing Code, Hands Off to AI Agents

Benoit Schillings, VP of Technology at Google DeepMind and former CTO of Google X, announced during a Wednesday keynote that he has stopped writing code after 45 years, delegating the task entirely to AI agents. He cited Gemini's ability to generate superior code as the turning point, describing it as 'superhuman syntax generation.' Schillings noted that roughly 80% of code on GitHub is now machine-generated, signaling the decline of the human code writer. He predicted that within a year, AI models like Gemini will generate code that no human will review, drawing a parallel to how developers no longer manually inspect compiler output. He also called for new programming languages designed to improve the security of AI-generated, agentic code.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Anthropic Research Pushes Spec-Driven AI Orchestration Over Prompt Engineering

Anthropic's engineering team has published research arguing that the infrastructure surrounding AI models matters as much as the models themselves, coining the term 'harness engineering' to describe this discipline. The research highlights recurring production failures including context degradation, over-ambitious single-pass agents, and unreliable self-evaluation in deployed systems. As teams scale from a handful of workflows to dozens, manually managed scaffolding breaks down and coordination overhead grows nonlinearly across multi-agent setups. The proposed shift moves away from bespoke prompt-based harnesses toward specification-driven contracts that define deliverables, verification criteria, agent assignments, and cost controls in structured but plain-English terms. Predicate Ventures' Blake Aber argues this declarative approach mirrors how Terraform and Kubernetes transformed infrastructure management, potentially compounding efficiency gains across entire product portfolios.

0 comments Read more at DEV Community