As AI takes over coding, Oracle rethinks how applications are built
That creates a trade-off IT leaders should examine closely. A compact application definition may be easier to audit than thousands of lines of generated code, yet operational confidence will depend on what remains visible outside the definition itself. Teams will still need answers to familiar engineering questions: how business logic is tested, how generated behavior is traced in production, how exceptions are handled, and how platform updates affect existing applications even if source definitions do not change.
There is also an architectural consequence. Once AI agents are optimized to emit a vendor-specific application language, the lock-in discussion moves upstream from runtime hosting to application design itself. For enterprise teams, the practical evaluation is less “can AI build apps faster?” and more “which parts of our portfolio are safe to express in a high-abstraction runtime?” Internal workflow apps, forms-driven systems and standard CRUD-heavy processes may fit well; highly differentiated systems with complex integration, performance constraints or bespoke control paths may not.
Before adopting this model broadly, practitioners should define review boundaries: what must stay declarative, what requires conventional code escape hatches, and how generated app definitions will be cataloged, tested and integrated into existing DevSecOps controls.
As vibe coding moves from novelty to everyday practice, enterprises face a new problem: Machine-generated software still has to be maintained, governed and understood by people.
Low-code platforms were built to speed up application development, and AI is now rewriting that mission. Oracle Corp. previewed its answer ahead of its annual conference, and the tipping point for software development has already arrived, according to Mike Hichwa (pictured), senior vice president of software engineering at Oracle.
“In 2025, the majority of code was human-authored, and in 2026 and forever afterwards, the majority of code is going to be machine-generated,” Hichwa said. “We’re probably not going back. What we’ve done to address this change is we invented our own language. We call it APEXlang.”
Hichwa spoke with theCUBE Research’s Dave Vellante at Oracle AI World, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed vibe coding and the need to maintain and govern AI-generated code. (* Disclosure below.)
Vibe coding meets enterprise governance
APEXlang is a highly abstracted application definition language that captures intent rather than implementation. Oracle introduced it with the APEX AI Application Generator and is publishing its grammar, skills and documentation in the public domain so coding agents such as Anthropic PBC’s Claude and OpenAI Group PBC’s Codex can generate it. Oracle is also working with OpenAI to train models to generate the language more efficiently, Hichwa explained.
“What you’re generating actually matters, because once you generate it, and you generate it 100 times or 1,000 times within your enterprise, then you’re going to need to maintain it. You’re going to need to look after it,” Hichwa said. “What people want with their portfolio of applications, they want consistency, they want governance, they want ease of understanding.”
Governance comes from the runtime, too. Generated APEXlang is interpreted by the APEX engine, which Oracle patches and updates monthly, so customers don’t have to revisit what they built. The same design is meant to curb AI slop, since a compact definition is easier to read than sprawling generated code, according to Hichwa.
“What we’re generating is much, much smaller, much more human understandable, so it is less sloppy because it is more abstracted,” Hichwa said. “It’s like a CliffsNotes. It’s not the full book.”
Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of Oracle AI World:
(* Disclosure: TheCUBE is a paid media partner for Oracle AI World. Neither Oracle, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)
As AI takes over coding, Oracle rethinks how applications are built
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.