The developer stories enterprise IT leaders actually read.
Volume 342, Issue 3 · 10 articles
Agent-run SOCs and machine-speed containment could reshape security roles the way AI reshaped development.
Google Cloud's 2025 State of AI-Assisted Software Development found 90% of developer respondents using AI and 80% believing it raised productivity, while an earlier Microsoft study measured 26% more tasks completed.
A March 2026 Federal Reserve Board working paper found robust evidence that annual coder employment growth is about 3% lower than pre-ChatGPT.
Gartner predicts 80% of organizations will evolve large engineering teams into smaller AI-augmented ones by 2030, with new roles including AI-governance specialists and context designers.
The article asks whether security follows, with agent-run SOCs, continuous vulnerability triage, machine-speed containment and humans directing fleets of defensive agents.
Vibe coding lets non-programmers build functional apps by describing ideas in natural language for AI.
Vibe coding represents a shift in app development where users describe their goals in plain English and AI models generate, execute, and refine the code accordingly.
Coined by researcher Andrej Karpathy in February 2025 and named Collins Dictionary's Word of the Year, this approach democratizes software creation for people without programming backgrounds.
Platforms like Lovable, Replit, and GitHub Copilot automate tasks such as database setup and debugging. While roughly 25 percent of Y Combinator startups used AI-generated code by mid-2026, experts caution that treating AI output as a first draft remains crucial for security and accuracy.
Comprehensive review of nine leading AI coding assistants, with GitHub Copilot rated best overall.
The article evaluates nine AI coding tools including GitHub Copilot, Cursor, Claude Code, and others across criteria like context awareness, reasoning capabilities, and workflow integration.
GitHub Copilot emerges as the strongest all-around choice for developers seeking AI assistance without changing their IDE setup, while Cursor excels as an AI-native editor and Claude Code specializes in autonomous complex tasks.
The market for AI coding tools is projected to grow from $4.3 billion in 2023 to $12.6 billion by 2028, reflecting widespread developer adoption.
AI-generated code requires rethinking code review and testing practices while unlocking developer creativity.
Phillip Mortimer argues that AI code generation has made all software write-only, prioritizing tests and disposability over readability.
He explains why traditional pull requests are broken and proposes automation of code reviews, extensive testing, and self-healing architecture.
As implementation becomes commoditized, developers should focus on creative problem-solving and architectural innovation rather than line-by-line coding.
AI-generated pull requests are overwhelming the senior engineers who built the review culture teams depend on.
The engineers drowning in review queues are not the AI skeptics, they are the early adopters who care most about code quality.
One study found 77% of engineers now spend less time writing code and more time reviewing AI output, while teams with high AI adoption merge 98% more pull requests and see review times rise 91%.
The deeper problem is cognitive: when a colleague writes code, intent travels with them through review, but AI-generated code arrives with the reasoning stripped out, leaving the reviewer to reverse-engineer intent from a diff.
That code also passes the eye test, reading coherently and handling the happy path while edge cases hide misaligned assumptions. The result is a review bottleneck that burns out exactly the people teams can least afford to lose.
Scanners flag severity but cannot infer business impact, so triage without context wastes scarce engineering hours.
A researcher scanning a 300-person B2B company found an internet-exposed database with weak authentication, rated critical and an obvious first fix. It turned out to be a resettable test database used to screen job candidates, holding no client data at all.
The episode illustrates why severity scores alone cannot drive remediation order: scanners have no way to infer the real cost of a compromise.
Security programs now absorb identity events, firewall logs, endpoint alerts, vendor feeds and threat intelligence, producing far more analysis data than a few years ago, and every item looks urgent. The piece argues for adding business context to triage, increasingly via external partners, so finite engineering hours go to findings that actually matter.
A survey finds 80% of developers describe AI as a dependence rather than an advantage, and managers reward heavy use.
The AI Coding Addiction Report surveyed more than 300 developers who use AI at least weekly, across seniority levels and tools including Claude Code, Gemini, Codex and Copilot.
Overall 80% described their relationship with AI as more like a dependence than an advantage, and respondents found it harder to give up AI coding tools than social media or video games. Some 43% reach the end of the workday but cannot tear themselves away from the reward loop, with developers reporting skipped breaks, meals and bedtimes.
The article frames the coding loop as a Skinner box, delivering frequent small rewards, and notes that managers are amplifying it by rewarding the heaviest users even when they ship code they do not understand.
With 41% of global code now AI-generated, standards must be explicit, available and difficult to bypass.
The article starts from a simple question for DevOps organizations: how does the AI know your standards, and what enforces them? One cited report says 92% of developers use AI coding tools daily in 2026 and 41% of global code is AI-generated, raising the stakes on whether that code delivers value or mounting AI debt.
Blumira's Mike Toole warns that generative AI will hand you the movie set of an app, with doors and windows that open and a clean demo but no input validation, auth boundaries or anything actually holding it up.
Directing coding agents requires the same skills as managing people: clear assignments, context, checkpoints and evaluation.
The industry long ago created Architect, Staff Developer and Chief Scientist roles so strong coders would not have to become people managers to advance, but seniority always meant dealing with more people anyway. Coding agents extend that dynamic to everyone.
Like humans, agents make bad decisions, write bugs and do things they were not asked to do, so they need clear instructions, wise guidance and the right working environment, plus supervision to stay on track.
Zide Software's Jim Bourke argues a coding agent is most useful when treated like an employee rather than an autocomplete tool: it needs a clear assignment, the right context, a safe place to work, checkpoints and a way to evaluate the result.
Australia's tax office deploys Copilot broadly not for ROI but for 'return on employee,' and a COBOL dev proved the point.
Australian Taxation Office CIO Mark Sawade told Gartner's IT Symposium that he is putting Copilot in the hands of as many administrative workers as possible to build the organizational muscle for AI, while deliberately keeping generative and agentic AI away from core services and the fraud detection systems the ATO has run on machine learning for a decade.
He does not expect obvious efficiency gains or ROI from the rollout, framing the goal instead as return on employee: developing the skills that later improve business processes.
The concrete evidence came from an internal hackathon where a COBOL developer won using GitHub Copilot, despite never having written a line of C# or anything in .NET.
More from Developer · All IT section newsletters · This week