In the early days of generative AI, AI-driven programming seemed to promise endless possibility, or at least a free pass to vibe code your way into quick wins. But now that era of freewheeling experimentation is coming to an end. As AI works its way deeper into the enterprise, a more mature architecture is taking shape. Risk-aware engineering, golden paths, and AI governance frameworks are quickly becoming the new requirements for AI adoption. This month is all about the emerging disciplines that make AI predictable, responsible, and ready to scale.

Top picks for generative AI readers on InfoWorld

What is vibe coding? AI writes the code so developers can think big
Curious about the vibe shift in programming? Hear from developers who’ve been letting AI tools write their code for them, with sometimes great and sometimes disastrous results.

The hidden skills behind the AI engineer
Vibe coding only gets you so far. As AI systems scale, the real work shifts to evaluation loops, model swaps, and risk-aware architecture. The role of AI engineer has evolved into a discipline built on testing, adaptability, and de-risking—not just clever AI prompts.

Building a golden path to AI
Your team members may not be straight-up vibe coding, but they’re almost certainly using AI tools that management hasn’t signed off on, which is like shadow IT on steroids. The best way to fight it isn’t outright bans, but guardrails that nudge developers in the right direction.

Boring governance is the path to real AI adoption
Big companies in heavily regulated industries like banking need internal AI governance policies before they’ll go all-in on the technology. Getting there quick enough to stay ahead of the curve is the trick.

How to start developing a balanced AI governance strategy
They say the best defense is a good offense, and when it comes to AI governance, organizations need both. Get expert tips for building your AI governance strategy from the ground up.

More good reads and generative AI updates elsewhere

Why AI breaks bad
One of the biggest barriers to corporate AI adoption is that the tools aren’t deterministic—it’s impossible to predict exactly what they’ll do, and sometimes they go inexplicably wrong. A branch of AI research called mechanistic interpretability aims to change that, making digital minds more transparent.

MCP doesn’t move data. It moves trust
The Model Context Protocol extends AI tools’ ability to access real-world data and functionality. The good news is that it acts as a trust layer, allowing LLMs to make those tool calls safely without needing to see credentials, touch systems, or improvise network behavior.

Anthropic says Chinese hackers used its AI in online attack
While details are scarce, Anthropic claims that Chinese hackers made extensive use of its Claude Code tool in a coordinated cyberattack program. The company says it’s working to develop classifiers that will flag such malicious activity.