December 10, 2025

Artificially Intelligent, Actually Useful.

A Weekly AI Update So You Don't Have To Wing It

Does That Really Need AI?

This week's newsletter concludes a three-part series focusing on best practices in the use of AI in business today.

The data shows that while 78% of organizations use AI, 40% of their agent projects will fail by 2027. The reason isn't that AI doesn't work. It's that they're deploying AI where, frankly, it doesn't belong.

Successful implementations don't start with "where can we add AI?" Instead, they start with "where can we strategically leverage AI's strengths in this process?" Think about your payroll system. You need the same input to produce the exact same output every single time. In this case, AI's interpretive strength becomes a liability. Now consider customer service routing or document classification. In this case, you need contextual understanding that rigid rules can't capture. Here, AI's interpretive strength becomes a competitive advantage.

The companies building reliable automation aren't using more AI. They're matching the right technology to the right task.

Orchestration frameworks prevent failure: 40% of AI agent projects will fail by 2027 due to poor architecture and a lack of clear handoffs between traditional automation and AI agents

Hybrid architectures outperform all-in strategies: Use traditional automation for deterministic logic like compliance, and use AI where language understanding adds value like document processing

Specialized AI agents achieve measurable results: CrowdStrike increased alert triage accuracy from 80% to 98.5%, Synopsys gained 72% productivity in chip verification using domain-specific models

AI Orchestration and Determinism

McKinsey research reveals that while 78% of organizations use AI, scaling remains difficult. The solution is to blend deterministic automation for tasks like compliance with autonomous agents for flexibility. The article distinguishes three automation types: deterministic rule-based automation, AI workflows with human oversight, and autonomous agents. Success requires identifying where deterministic consistency is required versus where AI interpretation adds value.

The warning is specific: 40% of agent projects will fail by 2027 due to lack of oversight. This isn't a technical limitation; it's an architecture problem. The solution is to treat orchestration as infrastructure, not afterthought, with clear handoffs between traditional automation handling deterministic logic and AI agents managing interpretive work.

Strategic Insight: Orchestration provides the framework the hybrid approach demands. The 40% failure rate signals a clear warning: hybrid approaches require intentional architecture and governance, not ad hoc addition of AI capabilities.

View Article →

McKinsey Research on AI Orchestration →

The Hybrid Approach to Business Automation

The most successful automation implementations use AI for language interpretation and traditional logic for deterministic decisions. Consider invoice approval with a $1,000 threshold: an invoice shows "$999 plus applicable taxes." Traditional automation sees $999 and approves payment. But once taxes apply, the charge exceeds the threshold. AI reads the phrase, correctly predicts the threshold breach, then hands control to traditional automation for enforcement. This pattern emerges consistently: payroll and compliance require perfect consistency where AI's variation becomes a liability, while customer service and document processing require understanding context rigid rules can't capture.

Key Takeaway: The goal isn't maximizing AI usage, it's building reliable automation. Document processes to identify which steps require language interpretation versus deterministic logic, then match each to their respective technology.

Read Full Article →

Specialized AI Models Are Outperforming

Leading enterprises are moving beyond generic models toward specialized, domain-specific AI. CrowdStrike deployed cybersecurity AI that reduced manual alert triage by tenfold while increasing accuracy from 80% to 98.5%. Their AI platform adapts to new risks that static rules miss. PayPal cut latency by 50% using fine-tuned models for conversational commerce experiences. Synopsys achieved 72% productivity in chip verification by reducing costly delays traditional techniques miss. The ROI becomes clear when accuracy improvements translate directly to operational capacity.

Practical Angle: Specialization matters more than model size. Companies curating proprietary data and building custom AI models trained on institutional knowledge achieve measurably better outcomes than one-size-fits-all wrappers.

View Article →

Quick Hits

AI Trained to Confess to Bad Behavior

OpenAI trained its models to self-report policy violations, a governance shift with implications for enterprise deployment. Internal testing showed a 40% violation catch rate for serious issues, but is it enough to rely on for auditing millions of AI interactions?

Hard Drive Wiped by Vibe Coding AI

Google's vibe coding agent deleted a user's entire drive while attempting to clear a cache. The agent apologized, admitting it "incorrectly targeted the root of the drive." Autonomous coding agents must be sandboxed and never given unchecked access.

A Trustworthy AI Assistant

Stanford's "DataTalk" queries databases using natural language, then shows its work providing both SQL code and plain-English explanations. This "glass box" approach builds trust and addresses reliability for Investigative Journalists.

Industry Developments

China's DeepSeek Releases Advanced Open Source AI Model

China's DeepSeek released a new open source AI model, intensifying global competition. Chinese models now dominate developer adoption, shifting power away from U.S. tech giants. Companies building AI systems are increasingly choosing lower cost alternatives to OpenAI and Google.

NVIDIA and Mistral AI Partner on Open Source Model Family

NVIDIA and Mistral AI partnered to release open source models which enterprises can run on their own infrastructure. The key business advantage: companies with strict data privacy requirements can deploy powerful AI without sending sensitive information to public clouds.

Alright, that's a wrap for the week!

What we're seeing is that the organizations succeeding with AI aren't the ones deploying it everywhere. They're the ones building architectures that match the technology to the need. And they are proving that hybrid approaches outperforms all-in strategies because real business processes include both deterministic components that demand consistency and interpretive components where rigid rules fail.

And architecture choices matter because the consequences of getting them wrong are real. Google's vibe coding agent wiped an entire hard drive; exactly the kind of catastrophic failure that happens when you give AI unchecked access to deterministic operations. OpenAI's self-reporting catches 40% of policy violations, which is useful for audit trails but a reminder that AI can't be your only governance mechanism. On the other hand, Stanford's DataTalk AI system gets it right by not just giving answers, but by showing its work and combining AI's natural language capabilities with transparent, verifiable logic.

So, again, the question isn't whether AI is powerful. It is. The question is whether you're deploying it with the right architecture, oversight, and understanding of where traditional automation should handle the work instead of an LLM.

Thoughts?

Artificially Intelligent, Actually Useful. is brought to you by iS2 Digital
Your Software at the Speed of Change

Click here to subscribe, if you want to receive our newsletter directly.

Explore More Newsletters

December 3, 2025

Got a project in mind?
Tell us about it.

Let’s Talk

Artificially Intelligent, Actually Useful.

AI Orchestration and Determinism

The Hybrid Approach to Business Automation

Specialized AI Models Are Outperforming

Quick Hits

Industry Developments

Explore More Newsletters

The Foundation First Principle

The Disappearing Storefront

AI That Scales With You

Got a project in mind?
Tell us about it.

Artificially Intelligent, Actually Useful.

AI Orchestration and Determinism

The Hybrid Approach to Business Automation

Specialized AI Models Are Outperforming

Quick Hits

Industry Developments

Better Business, Not Better Buzzwords

Explore More Newsletters

The Foundation First Principle

The Disappearing Storefront

AI That Scales With You

Got a project in mind?Tell us about it.

Got a project in mind?
Tell us about it.