Hanif Karimi

Still learning. Still building. Still questioning assumptions.

I’ve spent more than 20 years working across software engineering, architecture, cloud, platforms, identity, DevSecOps and large-scale technology transformation. Along the way I’ve had the chance to work on more than 100 enterprise projects, from hands-on engineering and solution design through to architecture and the leadership of large, multidisciplinary programmes, including nationally scaled digital services.

That experience has taught me something fairly simple: the difficult part of technology is rarely the technology alone. A solution has to work inside an organisation that already exists. It has to integrate with other systems, respect security and authority boundaries, survive production, make economic sense, and help someone achieve something useful.

AI doesn’t remove any of that. In many ways it makes it more important.

Why Contextual Agentic

The pace of progress in AI has made this an unusually interesting time to be building technology. Models can reason, generate software, use tools, retrieve knowledge and increasingly take part in workflows that used to need a person at every step.

But there is a long distance between an impressive demonstration and a system an organisation can responsibly depend on. That gap is what interests me.

I spend a lot of my own time building and testing ideas about agentic systems: context, models, tools, evaluation, identity, authority, evidence and governance. A question sits behind most of that work:

How do we get the benefits of increasingly capable AI systems while keeping people meaningfully in control of authority, evidence and outcomes?

I don’t think that question has a purely technical answer. It touches architecture, security, identity, data, operating models, economics, governance, organisational design and, in the end, human judgement. Those intersections are the part I find most interesting.

From experiments to systems

AI makes it remarkably easy to build something impressive. Making it dependable is harder. A prototype can work beautifully in a controlled demonstration and still fail when it meets incomplete context, changing data, ambiguous instructions, unreliable tools, security boundaries, cost constraints or unexpected human behaviour.

So I’m most interested in what happens after the demo:

• How do we know an AI system is actually working, and how do we evaluate it?
• What evidence should it produce?
• What authority should an agent have, and what should stay deterministic?
• When should a human approve an action?
• How do identity and permissions work when software begins acting on someone’s behalf?
• How do we observe these systems when they fail?
• And does the value justify the complexity?

Those are architecture questions. They are also organisational ones.

Building as a way of learning

I still build as well as advise. Staying close enough to the technology to test an idea, challenge an architecture, read the code and see where things actually break feels like an important part of technology leadership rather than a distraction from it.

AI has made that loop much faster. It’s now possible to explore architectures, build prototypes and test assumptions at a speed that would have needed a much larger team only a short while ago. But faster construction doesn’t remove the need for engineering discipline. If anything it raises it: code can become cheap while understanding, judgement and accountability stay expensive.

The Contextual Agentic Enterprise

This work eventually became a book, The Contextual Agentic Enterprise. It explores what happens when enterprise systems move beyond passive systems of record towards systems containing agents that interpret context, reason, use tools and take part in real organisational processes.

My interest isn’t in replacing people with autonomous software. It’s in how people and increasingly capable systems can work together while keeping clear boundaries around authority, responsibility, evidence and control.

The ideas keep evolving as the technology does, so I don’t treat the book, or this site, as the final word. They’re part of an ongoing investigation.

What you’ll find here

This site is where I share some of that work: practical architecture patterns, technical experiments, research, code, lessons from building, and thoughts about enterprise AI and agentic systems.

Some of it is deeply technical. Some of it is about strategy, governance, economics, organisational design or the changing role of technology professionals. Wherever I can, I want the work grounded in evidence, working systems and real engineering constraints rather than in AI hype.

I also expect some of it to change. The field is moving too quickly for anyone to pretend they have all the answers. When better evidence turns up, the thinking should change with it.

An open conversation

I don’t expect everyone to agree with everything here, and disagreement is useful. Architecture usually gets better when assumptions are challenged and other perspectives come into the discussion.

If you’re an engineer, an architect, a researcher, a technology leader, or simply someone trying to work out where AI and enterprise technology are heading, I hope something here helps you think, build, or ask a better question. And if you’ve reached a different conclusion, I’d genuinely like to understand why.

Independent views

Everything I publish here is my own independent professional perspective. It does not represent the views, policies or endorsement of any employer, client, partner, vendor or regulator, past or present, and none of them has reviewed it. The frameworks are educational design perspectives, not prescriptions; see the disclaimer.

Still learning. Still building. Still questioning assumptions.