Skip to content

Agentic AI Follows the Same Hype Cycle as SOA, Microservices and Cloud-Native, Warns Architect

An InfoWorld commentary argues that enterprises are repeating a familiar pattern with agentic AI: strong vendor hype, followed by higher costs and complexity than traditional systems. The author cites one project that cost roughly 20 times more than a conventional design.

2 min readBased on 1 source0 views
AI & ML

TL;DR

  • InfoWorld's author says agentic AI is entering the same adoption cycle that SOA, microservices and cloud-native architecture went through.
  • A regional distributor's agentic sales system reportedly cost roughly 20 times more than a traditional design and delivered about the same ROI.
  • The recommendation is to start from measurable business value and use agents only where the problem requires interpretation and adaptation.

A familiar pattern

In a commentary published by InfoWorld, enterprise architect author describes a project that started with a strong sales pitch. A large regional distributor wanted to automate parts of order intake, discount approvals, exception handling, customer communications and inventory checks. Vendors pitched agentic AI platforms that could make decisions, orchestrate workflows and learn from business context. The company chose an agentic architecture.

The build proved harder than expected. Each agent needed clean data, governed APIs, business rules, security controls, logging, test harnesses and human override mechanisms. Costs rose and security reviews slowed the work. The system ran in production, but six months later the company found that it cost roughly 20 times more than a traditional architecture would have, including supporting complexity, while delivering about the same ROI.

What happened to earlier architectures

The author argues that the pattern is not new. A technology emerges, early results look impressive, vendors package it and conferences promote it, until teams start from the trend rather than the business problem. He points to service-oriented architecture, microservices, cloud-native and older distributed object systems such as CORBA.

Each had real benefits. SOA encouraged service reuse, but enterprise service buses became chokepoints and governance grew heavy. Microservices allowed independent deployment, but many companies split ordinary applications into many services and took on distributed failure modes, data consistency issues and monitoring overhead. Cloud-native brought portability and elastic scaling, but many enterprises added clusters, service meshes and platform teams that their workloads did not need.

Where agentic AI adds cost

Agentic systems need model access, prompt management, tool orchestration, evaluation frameworks, audit trails, human-in-the-loop design and new operational skills. The author notes that they also bring nondeterministic behavior into processes that may need predictable results. For a developer, the practical meaning is that every agent adds a layer of testing and governance that a rules engine or plain API integration would not need.

The author says agentic AI fits best where a process needs interpretation, adaptation and semi-autonomous action. It is less suitable for stable, rules-based or heavily regulated processes that traditional logic can express. A workflow engine, rules platform or conventional application may deliver the same result at a fraction of the cost, and he argues that this difference should decide the architecture.

Why it matters

For developers and teams, the article is a reminder to define the business outcome and its measurable value before choosing an agentic design. Teams should compare agentic options with traditional workflow, rules and API approaches, and estimate the cost of governance, security reviews, testing and operations, not just the model calls. Architects who have lived through SOA and microservices projects will recognize the warning. Before adopting agents as a default, they need proof that simpler tools cannot meet the requirement. Readers who want background on cost trade-offs in cloud architectures can compare this with earlier discussions on our site.

Sources

This article was written by Blozhik from the reporting below. Read the originals for full detail.

InAgentic AI repeats same old enterprise adoption cycleInfoWorld
Was this useful?

Keep reading

Discussion · 0

Sign in to join the discussion. Liking does not need an account.Sign inRegister
Loading comments…
blozhikDeveloper news from across the web, explained. Every story links to its original sources.