Everyone wants you to move faster. You are the one who has to live with the decision.
Technology and AI decisions are getting larger, arriving faster, and becoming harder to reverse. The vendor has an answer. Your own team has an answer. The board wants progress. What is usually missing is an independent read of what is actually worth doing, what can wait, and what the organisation is genuinely ready for.
I'm Agron. I work with leaders when an important technology decision needs more than another opinion.
Phogen is where I keep that advisory work, the technical thinking behind it, and the problems that shaped how I make technology decisions.
What brought you here?
One of these is usually closer to the truth than a job title.
AI is already moving through the organisation. The question is whether anyone has decided where it should stop.
There are licences, experiments, people using tools nobody approved, and a vendor with a proposal. What is often missing is a decision about what a system may see, what it may recommend, and what it may carry out on its own. That decision is cheap to make now and expensive to retrofit.
Deciding what a system may doEveryone has a recommendation. You still have to decide.
The supplier is selling. The internal team has a preference and a reason for it. The consultancy has a programme. Each of them is partly right, and none of them signs the approval or lives with it afterwards.
A second view before you commitTechnology costs compound quietly. So do the decisions underneath them.
Cloud, licences, AI consumption, managed services, the integration nobody planned for. The spend is precise and easy to find. The return is usually a story somebody tells. The useful question is which part of the cost is genuinely unavoidable.
It works. Nobody is entirely comfortable explaining why.
An arrangement that has never failed can still be a liability: an unsupported system, a supplier who became structural, a workflow that lives in one person's memory. The risk stays invisible until the day it is not.
The organisation may already be using technology you have not decided to trust.
Confidential material moves into tools that were never approved, usually by capable people trying to do their jobs well. In a practice where confidentiality is the product, that is a professional exposure before it is a technical one.
How to use AI without giving up control of itYou do not need to become the technologist. You need enough clarity to decide.
Some decisions are too consequential to delegate and too technical to judge from a board paper. The job is to make the trade-offs legible, not to turn the board into engineers.
Why these decisions are hard
Advice is abundant. Accountability is not.
A technology decision of any size now arrives with several confident answers already attached to it. The supplier has one. The internal team has one, and usually a good reason for it. The consultancy has a programme. The security specialist wants more controls. They can all be partly right at the same time.
None of them approves the spend, explains it to the board, or runs the organisation that has to live inside the result. That part does not distribute.
What is usually missing is not another opinion. It is someone senior who is not selling the answer.
I have worked on these decisions from several sides of the table: infrastructure, security, data, product, AI, operations and executive leadership. That matters because the difficult ones rarely stay inside one of those boxes. A cost problem turns out to be an architecture problem. An AI problem turns out to be a question about who owns the data. Specialist advice is usually correct and usually incomplete.
Ways to work together
Four shapes account for most of the work I do.
Leadership for a defined period
Interim leadership and resetsA second opinion before an important decision
Board and principal adviceOne clearly bounded problem
Bounded mandates
Selected work
Three of the ten.
01
Trust and advisory, regulated financial services
Ninety days to retire an estate of legacy systems
Replacing a fragile collection of ageing systems before a failure made the decision for us.
Result
Finding a document fell from hours to seconds. A substantial share came off the annual licensing bill, because retiring the old systems retired what they cost to license. Most of the intended users were working in the new systems inside the first month, which is what actually decides whether a migration worked or merely finished. And recovery was not assumed. It was demonstrated, in a ransomware exercise.
02
AI systems, current work
An AI agent is privileged infrastructure
When AI can read company information and take action, its access deserves the same scrutiny as a powerful employee account.
Result
Security, identity and AI governance stop being three departments with three budgets and three different answers. Settled in advance, this is an afternoon's conversation. Settled afterwards, it is an incident.
04
Supervised machine learning, consumer services marketplace
A price without the site visit
Everyone in the marketplace was waiting on the same thing: somebody driving out to measure the job.
Result
Roughly a fifth came off what evaluating a job cost the customer. Offers could be issued immediately rather than after days of exchange, taking close to a fifth off the time spent producing them. A patent lawyer reviewed both methods and judged them novel enough to be worth protecting.
How I think about technology
Five positions, held long enough to be sure of them.
Start with the decision, not the product.
The question is almost never which tool. It is what the organisation is trying to be able to do, and what it will have to live with afterwards.
Make important information easier to reach, not merely easier to store.
Storage has been solved for years. Finding the right answer quickly, and knowing it is the right one, has not.
Give systems only the authority they need.
That has always been true of people's accounts. It became urgent the moment software started acting on its own.
Keep a way out before a supplier becomes too deeply embedded in the business.
Dependence is not a problem until it is the only option. By then it is expensive to discover.
Test recovery instead of assuming it.
Recovery that has never been tested is a theory. Most organisations own a document describing one.
Point of view
The safest AI system is not the one that does the least. It is the one whose authority somebody has actually decided on. There are six steps between a system that can only read and a system that can act on its own, and each one costs something different.
See how much authority an AI system is being given9Problems written up here
15Published pieces
6Questions I ask before trusting an AI system
Some things are deliberately absent.
You will not find client logos here, or a list of named organisations. Much of my work sits inside places where discretion is part of what they sell. I describe the problem, the reasoning and the outcome, and I leave the organisation out of it.
Background
The layers that now have to be understood together, I have run each of them: networks and infrastructure, security and risk, data science, delivery and platforms, product, and the executive layer where the question becomes which of them deserves the money.
That is why a board conversation and the architecture underneath it are the same conversation rather than two. University research, international development programmes, enterprise software, regulated financial services, and two things being built now.
More about how I got here
The useful time for an independent view is before the decision becomes difficult to reverse.
Two or three lines are enough: what is happening, what decision is coming, and when it matters. You do not need to know which kind of engagement this is. LinkedIn is the single route in, on purpose.