Start With A Safe Folder, Not A Live System
A practical operating note for introducing AI agents safely: start with a copied folder, safe material, clear project rules, no secrets, human review, and expansion only after the controls work.
Writing archive
Writing - Page 3
Older essays on agentic systems, governance, operations, incentives, and the work of making AI useful in real organisations.
144 pieces
Essays and notes
Published archive
Page 3 of 6
What I Would Do If I Were at University in 2026
A practical, optimistic note for students and universities: AI changes the cost of knowledge work, but the backlog of human need is still enormous.
Change Management for Long-Lived Agents
Long-lived agentic threads do not just need memory and context. They need a way to absorb change, challenge old instructions, and keep a useful history of why the rules moved.
Where Is the Productivity We Were Promised?
A practical challenge to SaaS vendors, public services, and leaders: if AI has made teams more productive, where are the better features, lower costs, faster services, and visible outcomes?
When Do We Need Judgement?
A practical note on why judgement engines should be called deliberately, not sprayed across every conversation, and how agents might use them when a decision actually matters.
Sharing Is a Language
A practical note on why SaaS systems still matter for collaboration, how they create a shared language around data, and what a common-understanding MCP could give agentic teams.
Your Context Is the Next Lock-In
A practical note on why the next AI lock-in may be your context, not the model, and how to start building a useful context store for agents.
What Do Humans Do While The Agents Run?
A reflective note on the dead time that appears while capable AI agents run: how humans switch focus, supervise output, rest, talk, walk, and redesign workflows around waiting.
Give The Agent A Desk, Not The Keys
A practical operating note for introducing office agents: give them a contained workspace, limited access, clear supervision, simple tasks, and enough time for people to become supervisors of their own AI.
Start At The Edge
A practical starting point for individual AI productivity: use a small sandbox, a second account, and one repeatable expense workflow before trying to transform the whole company.
Delegated, Not Abdicated
A practical note on using RACI to separate AI execution from human accountability when documents and decisions are produced with AI.
What Does an Advert Look Like for an Agent?
If agents start doing more of the choosing, the winning advert looks less like persuasion and more like inspectable evidence.
Who Teaches the Agents When the Forums Go Quiet?
A reflection on what happens to the open knowledge commons when agents, not people, are doing more of the asking and learning.
Vibe Coding Is Not the Problem. Skipping the Grown-Up Bits Is
A practical argument for treating vibe coding as an early system maturity layer, not the enemy of production engineering.
The Workday Is Measured in Tokens Now
A personal reflection on running out of tokens, choosing the right level of intelligence for the task, and why the future workday may be bounded by model budget as much as hours.
I Can Do Everything Now. That's the Problem
A personal reflection on what happens when the old blockers disappear and the real work becomes choosing, pacing, and staying human.
You Are Competing With Token Cost, Not AGI
A Friday thought on AGI, remote work, job risk, and why the first labour-market fight may be against token cost rather than raw intelligence.
Your Company Has Two Choices Now
A voice-led note on the choice now facing organisations: keep forcing work through controlled screens, or let people work naturally while agentic systems carry the rules and integrations around them.
Why Some Of My Articles Now Have An Agent Canon
A short, plain-English explanation of Agent Canon: why Tonywood.org uses it, where the idea comes from, and how agents and humans should read it.
If Agentic Can Delete Everything, Something Is Wrong
A public note to Agentic and operators on operational resilience, backup isolation, RTO, RPO, and why no single actor should be able to destroy the way back.
I Built My Own Website Because Control Matters
A short note on leaving hosted website constraints behind, rebuilding Tonywood.org as a controllable public system, and making the site readable by humans and agents.
A Call to All Managers Rolling Out POCs: How You Reduce Failures
Most AI proof-of-concepts fail after the demo. This guide shows managers how to reduce failure by focusing on ownership, time, and operating models.
Smart People, Dumb Data: The Hidden System That Keeps You Guessing
They recruit smart people, invest in analytics, and talk about evidence-based decision making. Yet when I walk into a large company, I often see the same pattern.
Why Your Company Website Should Become An AI-Readable Data Room (And Not Another Brochure)
This post came from a conversation I had at the Porto summit with a CICF member. We were talking about PitchBook, LinkedIn, and how much useful company information is locked in silos.