Why AI Agents Are Becoming the New Operating Model for Business: Key Takeaways from Miguel Ferreira’s (AWS) Presentation
On June 17, the Baltic Business Club organised an exclusive executive visit to the Amazon Web Services (AWS) office in Tallinn, bringing together business leaders to explore how artificial intelligence is evolving from a promising technology into a practical tool for marketing, analytics, and business transformation.
One of the event’s featured speakers was Miguel Ferreira, Senior Solutions Architect at AWS. His presentation focused on a fundamental shift taking place in enterprise AI: the move from standalone AI tools to intelligent agent-based systems that become an integral part of business architecture.
This article highlights the key insights from his presentation and addresses one of the most important questions facing organizations today: Why are so many companies still struggling to generate real business value from AI?
AI Agents vs. Traditional Chatbots
Ferreira argued that today’s biggest misconception is thinking of AI and particularly LLMs (large language models) as nothing more than a conversational interface. A language model alone has no understanding of a company’s operations, customers, or internal knowledge. Without that business context, even the most advanced model remains disconnected from the work that actually creates value.
To transform a language model into an AI agent capable of delivering measurable business value, organizations must build four core architectural layers for such system:
- Foundation model – providing reasoning and decision-making capabilities;
- Context engineering – connecting the model with internal company data;
- Long-term memory – allowing the system to retain user context and previous interactions;
- Ability to act – enabling the AI to use external tools, APIs, browsers, and enterprise applications.
What truly makes an AI system “agentic” is its ability to complete autonomous feedback loops: take action, evaluate the outcome and learn from mistakes.
As Ferreira explained:
“You can have artificial intelligence, but if it doesn’t work with your own data, it’s not particularly useful. What makes a system agentic is its ability to take action, use external tools, and then observe the results of those actions.”
Why Businesses Are Moving Toward AI Agents
The rise of AI agents is being driven by very practical business challenges.
Organizations are facing an aging workforce, the loss of institutional knowledge, and growing operational costs caused by repetitive support tasks.
Companies across the Baltic and Nordic regions are already using Amazon Bedrock to address these issues.
- KONE: Preserving Engineering Expertise
Elevator manufacturer KONE faced a familiar problem: experienced field engineers were retiring, while younger technicians lacked decades of accumulated maintenance knowledge.
- Instead of relying on lengthy calls to internal support teams, technicians now use a Field Technician application powered by AI.
By simply taking a photo of an elevator component, the AI agent searches thousands of technical manuals and instantly identifies the correct documentation, dramatically reducing troubleshooting time and improving service efficiency.
- Sun Finance: AI-Powered Identity Verification
Sun Finance faced significant challenges with its identity verification process due to the limitations of traditional OCR technology. Processing multilingual identity documents with different layouts often resulted in extraction errors, causing around 60% of loan applications to require manual review, increasing both operational costs and processing times.
To address these challenges, Sun Finance implemented an AI-powered identity verification solution using Amazon Bedrock. By combining large language models for intelligent data structuring with Amazon Textract and Amazon Rekognition, the company significantly improved document processing accuracy and automated much more of the verification workflow.
The results have been significant:
- Increased data extraction accuracy from 79.7% to 90.8%
- Reduced document processing costs by 91%
- Cut processing time from up to 20 hours to under 5 seconds
- Reduced the volume of applications requiring manual review
The Technology Behind Enterprise AI: AWS’s AI Stack
Building AI agents at enterprise scale requires much more than a powerful language model. AWS has developed a comprehensive AI technology stack, reflecting Amazon’s long-term commitment to AI innovation.
The stack consists of three primary layers.
- Infrastructure Layer
Using expensive GPUs for every AI workload is neither practical nor cost-effective.
AWS provides specialized hardware for different purposes:
- AWS Trainium for model training;
- AWS Inferentia for efficient inference;
- AWS Graviton (ARM) processors for general-purpose computing.
These systems are connected through AWS’s high-performance Elastic Fabric Adapter (EFA) networking technology, designed to support large-scale AI workloads.
- Data Layer
One of the biggest obstacles to enterprise AI is fragmented corporate data.
AWS addresses this challenge in two complementary ways. First, its Zero-ETL integrations provide managed data pipelines between popular data sources and AWS services such as Amazon S3 and Amazon Redshift, eliminating the need to build and maintain traditional ETL workflows.
Second, AWS natively supports open standards such as Apache Iceberg, enabling organizations to manage open table formats on Amazon S3 and query data using standard SQL while maintaining interoperability across analytics and AI workloads.
- Model Platform
Amazon Bedrock provides a unified API for accessing leading foundation models, including:
- Anthropic Claude
- Meta Llama
- Google’s open-source Gemma family of models
- OpenAI models running securely within AWS
- Amazon Nova
This enables developers to switch between models depending on the specific business use case without rebuilding their applications.
Security: The Foundation of Enterprise AI
As AI agents move from experimentation into the implemented production, security quickly becomes a top priority.
These systems interact with sensitive business information and communicate directly with customers, making governance essential.
AWS approaches security at multiple levels.
- At the infrastructure level, the AWS Nitro System encrypts data inside virtual machines, preventing unauthorized access—even from data center operators;
- At the application level, Amazon Bedrock Guardrails helps ensure that AI systems remain accurate, safe, and compliant by applying configurable safeguards.
As Ferreira noted:
“Once you move into production, security becomes much more important. Guardrails validate both the agent’s response and the context it’s using, checking whether the answer is grounded in your documents or whether the model is “hallucinating”.”
Bedrock Guardrails provides protection in the following key areas:
- Personal Data Protection: automatically detects and prevents the exposure of sensitive customer or employee information (PII);
- Content Filtering: blocks harmful, inappropriate, or irrelevant prompts and responses;
- Context Grounding: reduces AI “hallucinations” by verifying that responses are based on trusted internal knowledge.
- Automated Reasoning: helps validate responses against logical rules, improve reliability, and reduce the risk of hallucinations.
If an AI agent attempts to generate information that does not exist in the organization’s documentation, the response can be blocked before it reaches the user.
Key Takeaways: From AI Tools to Intelligent Systems
Ferreira’s presentation illustrates a broader shift in enterprise AI. Three strategic lessons stand out for business leaders:
- AI agents are becoming the new standard for automation. Traditional chatbots are no longer enough. Organizations increasingly need autonomous digital workers capable of interacting with enterprise systems and business data.
- Infrastructure determines economics. Successful AI initiatives depend on flexible infrastructure, specialized hardware, and open data architectures that allow organizations to scale efficiently without exponentially increasing costs.
- Trust depends on security. AI cannot become part of critical business processes without robust governance. Guardrails, data protection, and factual grounding transform AI from an experimental technology into a reliable, enterprise-grade business platform.
As enterprise AI continues to mature, competitive advantage will increasingly come not from simply adopting AI, but from building intelligent systems that can reason, act, learn, and operate securely within the business itself.
This article summarizes the core insights from the Baltic Business Club meeting. The content is for informational purposes and reflects the practical frameworks shared by the invited experts and club members.
