Return to site

🤖☕ The REALITY of AI in the ENTERPRISE

AI shouldn't replace enterprise software. It should live inside it.

· AI

🔸 TLDR

Enterprises already spent decades encoding workflows, rules, permissions, edge cases and business knowledge into deterministic software.

The opportunity isn't a giant AI agent running the company.

It's:

Deterministic software → AI → deterministic software

Use AI where the system needs intelligence: understanding text, documents, natural language or other messy inputs. Keep everything else predictable.

🔸 WHAT ENTERPRISE AI ACTUALLY LOOKS LIKE

▪️ ["If something can be deterministic, it should stay deterministic."]

Deterministic code is cheaper, repeatable, testable and reliable. AI needs a strong reason to enter the flow.

▪️ ["AI has to be a small piece of overall software, not the other way around."]

Don't rebuild decades of business logic around an LLM. Insert AI only where non-deterministic intelligence creates value.

▪️ ["Whatever is deterministic today must stay deterministic."]

A useful boundary is often exactly where humans currently interpret documents, text or ambiguous information.

▪️ ["We force AI to produce Java objects."]

Types, validation, permissions and security still matter. With non-deterministic components, these safeguards become even more important.

▪️ ["AI is like water. It will find a way."]

Defining the goal isn't enough. Architecture must constrain how the system reaches it.

🔸 JAVA, SPRING & ARCHITECTURE

The discussion with Andrey Santrosyan, James Ward and Josh Long also makes another strong point:

AI doesn't make traditional engineering obsolete.

We still need:

▪️ Architecture

▪️ DDD and abstractions

▪️ Idempotency

▪️ Testing and evals

▪️ Observability

▪️ Security

▪️ Human-in-the-loop controls

Andrey even predicts AI could eventually feel as mundane as another application dependency:

Need a database? Connect one.

Need intelligence? Connect that too.

🎥 Full conversation:

🔸 TAKEAWAYS

▪️ Don't replace deterministic workflows with probabilistic ones.

▪️ Put AI at the intelligence boundaries of existing systems.

▪️ Treat agents like untrusted inputs: type, validate and constrain them.

▪️ Architecture matters even more in the AI era.

▪️ Enterprise AI may ultimately be less about "AI apps" and more about AI inside apps.

#AI #EnterpriseAI #Java #Spring #SpringAI #SoftwareArchitecture #AgenticAI #GenerativeAI #SoftwareEngineering

Go further with Java certification:

Java👇

Spring👇

SpringBook👇

JavaFullstackBook👇