Ideas from the work

Thinking in public, grounded in practice.

Things I am learning, questioning, building and noticing at work and beyond it.

LinkedIn post

When a technology failure becomes a moral failure

Scale, automation and removal statistics are not an adequate defence when a platform enables and monetises a path to serious harm against children.

  • Child safety
  • Platform accountability
  • Technology ethics

LinkedIn post

People need room to fail

Hiring and workplace systems should distinguish poor effort from honest thinking, learning and thoughtful attempts that do not work the first time.

  • Hiring
  • Learning
  • Leadership

LinkedIn article · May 11, 2026

Success looks different in my 40s

A personal reflection on ambition, usefulness, learning, stability, and choosing work that fits the life being built now.

  • Career
  • Learning
  • Perspective

LinkedIn post

Understand the work before automating it

Why useful AI systems begin with how skilled people handle intent, exceptions, validation, and judgment, rather than with a model choice.

  • Applied AI
  • Automation
  • Human judgment

LinkedIn article · May 25, 2025

Natural language is not a reliable specification

An experiment in synthetic-data generation found that moving one decision from the prompt into code made the output more precise and the workflow faster.

  • Prompt engineering
  • Data quality
  • Experimentation

LinkedIn post

Self-service agents still need ownership

Low-code creation is easy; visibility, testing, support, lifecycle control, and safe retirement are the harder enterprise design problem.

  • AI governance
  • Low-code
  • Enterprise systems

LinkedIn post

Context is part of the system

Translation validation shows why the right context often comes from structure, including the question, audience, subject and market, and why more context is not always better.

  • Context engineering
  • Data quality
  • Validation

LinkedIn post

Start with the problem, not the syntax

A personal reflection on declining a Python cookbook and watching AI change how a new generation begins learning to program.

  • Python
  • Learning
  • AI-assisted development

LinkedIn post

AI gets useful at the file-structure level

Experiments with DOCX editing illustrate the difference between giving instructions and working directly with Python libraries and document XML.

  • Agentic AI
  • Document automation
  • Python

LinkedIn post

Coding agents need a careful eye

A lighter note on the mix of delight and vigilance involved when powerful coding agents solve one problem and surprise you somewhere else.

  • Coding agents
  • Software practice

LinkedIn article · Apr 22, 2025

How a perfect curry became my AI architecture

A kitchen metaphor for breaking expert work into specialist roles, clear hand-offs, measurable standards, and responsible orchestration.

  • Agentic AI
  • Systems thinking
  • Orchestration

LinkedIn article · Apr 13, 2025

Designing transformation for agents, not only interfaces

An early argument for code-first, API-driven systems that expose clean logic and data to intelligent agents rather than treating another interface as the final transformation.

  • Transformation
  • Agentic systems
  • Architecture

LinkedIn post

Restrictions can create new competitors

Limiting access to frontier models may accelerate investment in local models, infrastructure and technological independence outside the United States.

  • AI sovereignty
  • Competition
  • Infrastructure

These summaries lead to the original writing on LinkedIn.