Your AI memory problem is a permissions problem
Every agent memory pitch is about recall. What stopped me building one company-wide was access control, and the systems that already solve it are the ones you're already paying for.
I have spent 28 years building software through every major shift the industry has been through, from the early web to the AI revolution. Today I am CTO at Juice, a London fintech, and I write about what leading engineering teams through change actually involves.
Every agent memory pitch is about recall. What stopped me building one company-wide was access control, and the systems that already solve it are the ones you're already paying for.
The most dangerous output an automated system can produce is a clean one. What I learnt when every scheduled job I relied on died quietly, and the checks I now use to tell 'nothing to do' from 'nothing worked'.
I started in software when the web was new, and I have been building and leading engineering teams ever since. Each wave changed what was possible and what customers expected. Each one also rewarded the same habits: clear thinking, honest measurement and teams that are trusted to own their work.
I am currently CTO at Juice, where I lead technology for a lending platform. Before that I worked with and consulted for many blue-chip UK and global companies.
These days most of my attention goes on the newest shift: what it means to run engineering when AI agents write a growing share of the code, and how to keep quality, security and judgement intact while you do it.
In the order they arrived.
Building for the browser when a website was a novelty and most pages were made by hand.
Social, user-generated and API-driven products, when software turned into a conversation.
Moving workloads from data centres to elastic platforms, and rethinking how systems are built and run.
The smartphone put a computer in every pocket and made consumer-grade experience the standard for business software too.
Lending, payments and financial data, where reliability, security and regulation are part of the product.
AI agents that write code and run operations, and the controls you need before you can trust them.
Strategy, architecture and engineering teams that deliver what the business actually needs.
Scalable platforms on Google Cloud, legacy modernisation and migration.
Lending systems and financial data, with the security and audit trail a regulated product needs.
Putting AI agents to work in software delivery and operations, with controls that hold.
If something I have written is useful to you, or you see it differently, I would like to hear about it.