From Code to Intelligence: The Rise of the Full-Stack AI Engineer - Adam Stephensen - NDC Oslo 2026
Summary
The talk outlines the evolution of software development into the realm of Full-Stack AI Engineering, arguing that AI project failure is rarely due to model quality or code deficiency, but rather due to systemic issues in the surrounding ecosystem. Success requires orchestrating intelligence across diverse layers—including low-code platforms, pro-code agents, data platforms, and governed systems. The speaker emphasizes that the modern developer's role is shifting from writing code to selecting and integrating the right tools and managing the data layer to ensure scalable, trustworthy, and compliant AI deployments.
Key takeaways
-
AI Success is an Ecosystem Problem
22:20
AI projects fail not because the model is inadequate, but because the surrounding infrastructure and processes are not built correctly. Success requires addressing the entire system, not just the AI component.
-
The Full-Stack AI Engineer Role
18:40
The developer's role is shifting from writing code to orchestrating intelligence across low-code and pro-code agents, data platforms, copilots, and governed systems.
-
Data Layer is the Primary Challenge
36:30
The most common failure point is the data layer. Challenges include poor data quality, lack of semantic layers, or data being inaccessible, making it impossible for the AI to answer complex questions.
-
Governance Must Be Shifted Left
53:20
To achieve scale, governance and security cannot be left until the end of the project. Implementing 'Trust Ops'—embedding governance, risk, and compliance into the development and operational cycles—is critical for production readiness.
Technical details
-
AI Architecture Evolution
1700s
The architecture has moved from traditional three-tier web apps (Front-end, API, Database) with simple RAG agents to complex, integrated systems. Modern systems leverage an 'IQ' layer (e.g., Work IQ, Foundry IQ) that acts as an abstraction over various data sources (SharePoint, Teams, OneDrive, etc.).
-
Interoperability Standards
1440s
The Messaging Communication Protocol (MCP) is highlighted as a game-changer, enabling solutions to share resources and plug together across different platforms (e.g., Salesforce, SAP, AWS, Google).
-
Data Platforms and Agents
1640s
Platforms like Microsoft Fabric and Databricks allow users to create AI agents directly within the data platform, eliminating the need for developers to write custom code to connect structured data to AI services.
-
Advanced AI Capabilities
2300s
Foundry IQ enables 'retrieval planning,' allowing a single complex, multi-part question to be broken down, solved in parallel across multiple data sources (e.g., Fabric IQ, Web IQ), and synthesized into a single response.
-
Testing and Reliability
3300s
To ensure trust at scale, developers should treat AI applications like traditional software by implementing 'plain language unit tests' or 'assertions.' This involves defining expected inputs, outputs, and acceptance criteria for every functional requirement.
-
Security Guardrails
3450s
For high-risk applications (e.g., finance), deterministic guardrails should be implemented in the deterministic part of the application (the API/tool call) rather than relying solely on the non-deterministic LLM output. The Agent Control Standard (ACS) is recommended for this.
Mentioned resources
- Microsoft AI Ecosystem
- Foundry IQ
- Microsoft Fabric
- Messaging Communication Protocol (MCP)
- Agent Control Standard (ACS)
Channel & topics
Watch on YouTube · Back to latest
This independent, AI-assisted summary is provided for commentary and informational purposes. It may contain errors or omit important context. Please watch the original video for the creator's complete presentation. Video, thumbnail, and related copyrights belong to their respective owners.