Blog
Thoughts on technology, infrastructure, and AI from 25+ years in the field.
Recent GitHub work shows why upgrades, clearer failure handling, and handover notes are now leadership concerns for firms running AI, automation, and client-facing systems.
Serious AI pilots need a rehearsal environment with realistic data, clear gates, fallback paths, and review controls before they touch live operations.
If you want an AI assessment or advisory funnel to win trust with serious buyers, the operating model has to be designed before the launch page goes live.
A weekly note on dependency drift, governance cadence, and the small fixes that make releases less stressful.
A weekly GitHub roundup showing how maintenance, documentation, and governance work are increasingly what make the product trustworthy.
A useful AI stack is not a collection of fashionable components. It is a system with clear roles, observable behaviour, and operating rules that make the parts work together.
This week's GitHub activity was dominated by production hardening: removing committed secrets, building a security scoring engine, overhauling cron reliability, and making model routing cost-first. Here's what happened and what I learned.
AI agents are no longer just answering questions — they are taking actions on your systems. Without audit trails, you cannot reconstruct what happened, prove compliance, or improve the system.
AI agents are moving from demos to production work. The organisations getting value from them are treating them like a production team, not a single chat window. Here is what that looks like.
Technology debt rarely kills a transaction outright. More often it weakens confidence, creates leverage for price chips, and makes management look less prepared than it should.
Microsoft 365 is the backbone of most UK businesses. But the default tenant configuration is not secure enough for a regulated organisation. Here is what to check and fix.
Connecting AI agents to business systems without a control plane is like running a data centre without monitoring — until something breaks, you have no idea what happened or how to fix it.
AI pilots are accelerating across UK SMEs. Without senior security leadership in place before they scale, organisations are exposing themselves to risks that are far cheaper to address upfront.
A Fractional IT Director gives a growing business senior technology leadership without the cost or timing risk of a full-time hire. The value is not advice alone. It is accountable direction.
AI agents become useful when you run them like a team: clear roles, explicit handoffs, and someone accountable for the result.
The biggest lesson from building AI systems was not about prompting. It was that decades of infrastructure, security, and delivery experience became more valuable once the pace increased.
The interesting part is not the automation stack itself. It is what happens when the stack becomes the thing that ships value.
I exported my own AI session history to see what it said about my working style, habits, and blind spots.
2024 was the year the tooling stopped being a novelty and started acting like infrastructure.
Multi-agent systems are powerful, but they need permissions, logging, and review before they touch production work.
A practical way to keep AI workflows under control: explicit budgets, human approval, and a nightly review loop.
The most useful way to think about AI governance is not as a separate discipline, but as an extension of the same zero-trust assumptions that already make sense in security architecture.
AI inside Microsoft 365 is only useful if the tenant is already under control.
When the AI tooling ecosystem exploded, I started pushing more of my experiments into a local homelab.