Understanding the system
AI Agents: What They Do and Where They Fit
Understand model-directed workflows, tool access, the surrounding AI harness and the limits that matter before deployment.
Read the articleIdeas grounded in engineering.
Practical writing on AI systems, the controls around them and the work of bringing them into a business.
Insights
27 articles
Human-in-the-loop AI: who should approve what an AI agent does
Decide which agent actions need human-in-the-loop AI, who should approve them and how to enforce approval through identity controls.
Choosing an AI Engineering Partner for Regulated Work
Questions to ask about scope, data access, evaluation, professional review and operating responsibility before selecting an AI partner.
RAG vs Fine-Tuning: Choose the Right Intervention
A practical comparison of retrieval and fine-tuning, including evaluation, data handling and when the two approaches belong together.
Closing the Reliability Gap in AI Systems
Why plausible answers need source checks, software validation, observable outcomes and an operating process around them.
Finding the Real Cost of an AI Workflow
Measure model calls, retrieval, retries, tool activity and human review to understand cost per completed task.
Production AI: Design for the Failure Paths
How to handle incomplete tasks, uncertain external writes and operational recovery when an AI prototype becomes a working system.
Where Automation Ends and AI Helps
How fixed workflows and AI interpretation can work together without making every business action depend on a model.
Taking an AI Pilot into Production
A staged approach to integration, acceptance testing, rollout and operating ownership without treating a pilot as proof of readiness.
Managing AI Requirements Across Jurisdictions
How to maintain a common engineering control set while assessing the laws and operating requirements that differ by organisation and use case.
EU AI Act: A Practical Scoping Guide for AI Projects
Understand the organisation’s role, use-case classification, current application dates and the evidence an engineering team needs to prepare.
Designing Model-Agnostic AI Applications
How to separate model interfaces from business logic while testing the real limits of portability and provider changes.
When On-Premise AI Makes Sense
A practical comparison of private and hosted AI, covering data flows, operating costs, security responsibilities and deployment constraints.
AI-Assisted SAP Modernisation with a Clear Review Path
How to use AI during SAP modernisation while preserving clean-core decisions, supported interfaces and release evidence.
Define the Work Before the AI Architecture
How to turn an AI initiative into a bounded project with requirements, data decisions, acceptance evidence and operating ownership.
Understanding Custom SAP Code Before Changing It
Use dependency, usage and test evidence to guide AI-assisted analysis of custom ABAP without treating generated code as verified business logic.
Managing Context in Enterprise AI Systems
How to select, preserve and validate the information an AI workflow needs without confusing a context window with durable memory.
AI Agent Risk: Follow the Permissions and Actions
How tool access, untrusted content and uncertain external outcomes change the risk of model-directed workflows.
Designing Human Approval into AI Workflows
How to make approval a specific, inspectable decision rather than a generic checkbox after a model response.
Connecting AI to Existing Business Systems
How to integrate AI with established software through narrow interfaces, explicit data ownership and tested recovery paths.
Security Controls for Private AI Deployments
Private hosting changes the risk boundary. It still needs access control, egress review, artifact integrity and application-side safeguards.
Why Data Quality Still Matters in AI Workflows
How inconsistent records, missing definitions and stale sources affect AI systems, and how to make those failures visible.
Why AI Projects Stall Before Production
How unclear ownership, weak acceptance criteria and missing integration work can prevent an AI pilot from becoming useful operational software.
An AI Dictionary for Practical Engineering Decisions
Twenty-five AI terms explained through what they mean, what they change and the limits to keep in mind.
AI Agents: What They Do and Where They Fit
Understand model-directed workflows, tool access, the surrounding AI harness and the limits that matter before deployment.
PIPEDA and AI Agents: Make the Data Flows Visible
Map personal information across prompts, retrieval, tools and logs so privacy obligations are assessed against the real AI workflow.
RAG: What Retrieval Can and Cannot Solve
How retrieval-augmented generation connects models to business knowledge, and where access, freshness and answer quality can still fail.
Assistants, Copilots and Agents: What Actually Changes?
A clear way to compare AI tools by their access, actions and human responsibilities rather than by product labels.