[BDD 2025 - Full-Stack Development] Agentic AI Architecture: Redefining System Communication

Agentic AI Architecture:
Redefining System
Communication
Donnie Prakoso
Principal Developer Advocate
Amazon Web Services

Education:
Donnie Prakoso
Work Experiences:
• Principal Developer Advocate
at Amazon Web Services
● Universitas Indonesia, Master of
Science
● Universitas Indonesia, Bachelor
of Information System

Helping developers
write code faster
2023
Completing
developmenttasks end-
to-end with human in
the loop
Generating larger
pieces of code and
answering questions
AI is changing software
development
AUTO-COMPLETE 2024 ASSISTANTS 2025 AGENTS
FUTURE
Influencing our
architectures and
agentic communication?
2026 WHAT’S NEXT?
PAST PRESENT

Choosing compute options
AWS
Lambda
AWS
AppRunner
Amazon
ECS
Amazon
EKS
AWS Fargate
Container Image

Service
A
Service
B
Service
C
Service
D

Boxes and
Lines:
System
designs
Service
A
Service
B
Service
C
Service
D
Service
A
Service
B
Service
C
Service
D

System architecture is
fundamentally defined by how
its services communicate.
This is not a quote, but the layoutlooks good.
“

Most commonly used
patterns:
● Choreography
● Orchestration
● Synchronous request-
response
• E-commerce platform
• A customer browses an online store, adds an item
to their cart, and completes a purchase.
• Video streaming service
• A user streams a video on their smart TV, and the
service recommends similar content.
• Financial services and banking
• A customer transfers money from their checking
account to another user.
Scenario: End-to-end business workflow

e.g. HTTP connection
Top
Bottom
Top
Bottom
Synchronous request-response pattern w/ APIs
Synchronous request-response pattern
Immediate response expected Requester waits till response is received
A
B
Requester Responder
12
3
9
6
Conversation pattern Common pattern for APIs

Synchronous integration:
Scenario order/payment
Amazon
API Gateway
Order service
Client
Invoice
service
Fulfilment
service
Forecast
service
SYNC SYNC (2) SYNC
<DEMO>

Receiver
TopLeft1
BottomLeft1
Asynchronous point to point
pattern w/ queues
Queues decouple sender and receiver by location and temporal dependencies
Sender C B A Queue C B A

TopRight1
BottomRight1
TopLeft1
BottomLeft1
Asynchronous request-response
pattern w/ queues
Single consumer pattern
Logical channels decouple location Queues minimize availability dependency
Message passing decouples interaction style and data format dependencies
Request-Queue
Response-Queue
Requester Responder
A A
B
12
3
9
6
B
A A
A
B B B

TopLeft1
BottomLeft1
Asynchronous publish /
subscribe pattern w/ topics
Topics fan out messages while queues act as buffering load balancers
Publisher
Subscriber
Subscriber
Subscriber
C B A
C A
B
C A
C A
B
B
Topic

event
[i-’vent] noun
A signal that a system’s state has
changed
Event is the language

Serverless is much more than compute
AWS
Lambda
AWS
Fargate
Compute
Data stores
Amazon Aurora
Serverless
Amazon Simple
Storage Service
(Amazon S3)
Amazon
DynamoDB
AWS
AppSync
Amazon
API Gateway
Amazon Simple
Notification
Service
(Amazon SNS)
Amazon Simple
Queue Service
(Amazon SQS)
AWS
Step Functions
Integration
Amazon
EventBridge

Asynchronous communication:
Scenario order/payment <DEMO>

Workflow Integration with AWS Step Functions
• Is built using a state machine
• is composed of steps called states
• Is written using Amazon States
Language or ASL
• Think of it as the workflow assembly language
• Can be used to orchestrate multiple
AWS services
Workflow orchestration:

How does GenAI influence
our architectures?

Something I discovered few months ago

From improving to transform
Create Content
outline.md
Create Workflow
workflow.md
Presentation
index.html
/css
/js
/images
Overall process
What I’m doing
Amazon Q CLI

Autonomous software systems that
leverage AI to reason, plan, and complete
tasks on behalf of humans or systems
AI Agents

Strands Agents
Strands Agents is an open source
python SDK for building agents
using just a few lines of code
< Hello, World! >

Provides agents standard access to an
expanding list of accessible tools
• Provides a standardized way to connect AI
models to different data sources and tools
• USB-C port for Agentic AI applications
• Open-source protocol developed by
Anthropic
Model Context
Protocol (MCP)

Agentic Communication
Evolution of service communication
MCP Strands
Microservices
+

Agentic
communication
proof of concept

Amazon Bedrock AgentCore — In Preview
Enhance with tools
and memory
Deploy securely
at scale
Monitor
AgentCore Runtime
AgentCore Identity
AgentCore Browser
AgentCore Code Interpreter
AgentCore Memory
AgentCore Gateway AgentCore Observability

Agentic
workflow proof
of concept

So, what’s next?
• To support business, you build systems — to build systems, you can use GenAI
• Foundational skills remain critical:
• Distributed system principles still apply — agentic workflow is just an abstraction layer
• Usage plans help throttle and gate agent requests
• Idempotency prevents duplicate operations from non-deterministic agents
• Deterministic inputs are required to handle non-deterministic AI outputs
• Service discovery and registry patterns need reimagining for agent communication

The future of microservices isn't just about better
APIs - it's about services that can think, reason, and
communicate intelligently through AI agents

GenAI can write your code and run your workflows,
but it can't replace your understanding of why the
system needs to exist in the first place.

Thank You
https://donnie.id @donnieprakoso
Donnie Prakoso
Get in touch
donnieprakoso
go.donnie.id/youtube

[BDD 2025 - Full-Stack Development] Agentic AI Architecture: Redefining System Communication

More Related Content

Similar to [BDD 2025 - Full-Stack Development] Agentic AI Architecture: Redefining System Communication

More from Wintari Yasmin

Recently uploaded

[BDD 2025 - Full-Stack Development] Agentic AI Architecture: Redefining System Communication