The Agentic Revolution

More clarity upfront. More quality downstream.

The AI Evolution: From Prompt to Agent

Prompt AI
Q1 2023
One question, one answer.
The human does everything.
AI only responds.
Oct 2023
Comelit adopts Copilot «It helps, but today it doesn't write code for us.»
Context AI
Q1 2024
AI understands context.
RAG, memory, grounding.
Smarter output, still reactive.
Agent AI
Q1 2025
AI acts autonomously.
Plans, decides, executes.
Humans supervise intent.
Oct 2025
Comelit starts adopting Agentic AI «It's starting to write code with us.»
What's
next? Autonomous AI Artificial General-Intelligence

The Agentic Layer: A New Management Architecture

1 Human Strategy & Intent Goals, ethics, strategic direction
delegates
2 Agentic Layer Autonomous orchestration & execution
accesses
3 Core System & Data Infrastructure & data foundation
→
Example agents
Coordinator agent Aligns priorities, routes work and combines specialist output into one flow.
Planning Agent Frames the task and prepares the next decision.
Coding Agent Builds and refines the implementation work.
QA Agent Checks quality, risk and release readiness.
Service Agent Handles domain requests, follow-up and escalation paths.

Management shifts from direct execution to governed orchestration.

The Agentic Layer: A New Management Architecture

A visual model for the operating flow.

Alternative agentic architecture visual showing human orchestration, multiple specialized agents, internal knowledge and external tools

Management shifts from direct execution to governed orchestration.

The Voice of Leaders

CODE REVIEW BY AI
“Agent usage in code review is very high”
CODE WRITTEN BY AI
“Maybe 20, 30% of the code inside of our repos today is written by software.”

Agentic Orchestra

Humans lead, four specialized agents, one connected delivery flow.

GrandStream 1953 example
Agentic workflow with planner, builder and tester connected to Comelit systems, Jira and delivery tools

Spec-Driven Delivery

Clearer output starts with a shared artifact.

Operating flow

Initial Brief

The request starts rough and still open to interpretation.

Clarifying Questions

Ambiguity is surfaced before the technical discussion starts.

Clear Plan

Intent, constraints and success criteria are now explicit.

Why it works

Better Specs Less ambiguity before work starts.
Better Boundaries Clear ownership for drafting, approval and execution.
Better Evaluation Small reviewable steps keep quality measurable.
One Clear Plan PM, application work and QA use the same reference.

One clearer plan. Fewer downstream surprises.

Pullreviewer

An agentic tool for code review.

How it helps
A code change arrives The team proposes an update and prepares it for review.
Pullreviewer reviews it The agent highlights what deserves attention before the team approves it.
The team decides faster Reviews become easier to read, easier to discuss and faster to close.

Same standards, same review format, less back-and-forth for the team.

Why it matters

It reads each proposed change The first review arrives immediately and points people to the parts that matter.
It gives consistent feedback Different teams see the same review style instead of fragmented comments.
It shortens the path to approval Less time interpreting feedback, more time deciding what to do next.

A clearer review flow for every software change.

Live Demo

A visible workflow from brief to qualification

GrandStream 1953 as a concrete adoption case

Not a promise. A visible end-to-end workflow.