AI That Works: A Pragmatic Guide to Real-World AI Systems

Beyond the Hype

At the Innovation Discovery Center, we've seen one pattern repeat itself: for every exciting new AI headline, there are a dozen implementation challenges that go unaddressed.

Theoretical breakthroughs make headlines. But working systems change industries.

That's why we're launching AI That Works—a new content series for leaders, engineers, and innovators who care about what's deployable, adaptable, and impactful today.

For example, a large multi-state healthcare system implemented an architecture like what we'll cover in our second edition, resulting in a 40% reduction in documentation time for nurses while improving clinical note quality and accuracy. We'll explore the key decisions that made this possible while protecting patient privacy and ensuring HIPAA compliance.

What Makes This Series Different

Some implementation guides offer generic best practices or vendor-specific documentation that serves marketing goals, our series will provide vendor-agnostic, architecture-level analysis based on real production systems. In addition to identifying what components can be used we’ll discuss how they connect, why certain design choices matter, and where common failure points emerge.

Each analysis will include a standardized evaluation framework—developed in collaboration with our IDC China team led by Xinghan Yan—that quantifies reliability, scalability, and maintainability factors often overlooked in typical implementation content.

Pressing AI Challenges in 2025

In this series we’ll tackle crucial AI challenges facing organizations in management, technical implementation, and operational deployment as of 2025:

• AI System Fragmentation & Toolchain Complexity: "We have too many disconnected models, tools, and data sources."

• Lack of Long-Term Memory and Context Awareness: "Our AI doesn't remember what happened yesterday."

• Hallucination, Verifiability, and Trust: "We can't trust what the model outputs, even if it sounds right."

• Model Adaptation vs Generalization Tradeoffs: "Do we fine-tune for our domain, or stay general for flexibility?"

• Scalability and Cost of Inference: "Our AI works, until usage spikes or budgets tighten."

• Evaluation Without Benchmarks: "We don't know how to measure real-world performance."

Two Tracks, One Purpose

To meet the needs of our diverse community, each post is split into two focused tracks:

Executive Track	Technical Track
Understand how this system solves real problems and drives ROI	Learn how it works, what tools it uses, and how to build or adapt it
See where this fits into your organization's strategy	Get a breakdown of models, memory, retrieval, and architecture
Ask the right questions of your team or vendors	Apply scoring frameworks to assess technical fitness
Focused on deployment, cost, safety, and long-term readiness	Focused on modularity, toolchains, scalability, and performance

Why This Matters Now

AI's rapid evolution has created a gap:

• Between what's possible and what's operational

• Between AI theory and engineering reality

• Between demo prototypes and scalable deployments

Researchers like Haifeng Wang have already shown the way forward by focusing on memory, infrastructure efficiency, and systems that scale to society-level use cases. Their work exemplifies the IDC mission: bridging research and reality.

Now we bring that same clarity and pragmatism to you.

What to Expect

In each upcoming post, we'll:

• Evaluate a real-world AI system diagram using our scoring framework

• Explain how the design addresses (or fails to address) key challenges like memory, safety, and scalability

• Offer use-case examples and organizational takeaways

• Recommend improvements grounded in 2024–2025 best practices (like LangGraph, OpenDevin, DSPy, etc.)

• Provide action prompts for both executives and engineers

Who It's For

• Executives and policymakers looking for actionable insight into AI strategy

• Product and engineering leaders designing next-generation platforms

• Researchers and builders focused on AI safety, memory, context, and modularity

• Healthcare and public sector teams trying to operationalize AI under pressure

  
  

Join the Journey

If you're ready to move beyond the noise and focus on what actually works, this series is for you.

The first edition drops soon, featuring AI Systems Under Pressure: Managing the Next Wave of Deployment Challenges.  A China–U.S. Perspective on Technical, Operational, and Governance Constraints in Enterprise AI.

Stay tuned.

Stay pragmatic.

Let's build AI that works.

Innovation Discovery Center

Grounded Innovation for a Smarter World