Workflows & transformation
I start by reconstructing the work as it actually happens: where decisions stall, where people compensate manually, where the documented process stops matching reality, and which failure is expensive enough to matter. The intervention comes after the operating truth is clear.
Follow the work end to end. Find the shadow process, the workarounds, the missing information, and the places where the documented workflow stops describing reality.
Separate the visible problem from the system causing it: a bad handoff, capacity mismatch, unclear ownership, weak data, broken incentive, or delayed decision.
Decide what actually deserves technology. AI, automation, workflow redesign, better information, tighter governance — or no new tool at all.
Define system behavior, human ownership, exception paths, permissions, inputs, outputs, and what success has to look like in the real environment.
Test with real users, expose edge cases, resolve friction, and keep adjusting until the new way of working can survive a normal bad day.
Measure the operating consequence. Faster matters. Safer matters. More capacity, less rework, recovered revenue, better access, or a better decision matters more.
How I view AI
I do not think the most interesting future of AI is a better answer box. Intelligence becomes more meaningful when it can remain present across time, understand context as it changes, and develop better judgment instead of beginning from zero every time someone asks a question.
I do not think people should have to continually stop what they are doing, formulate a prompt, and explain the world again. The more natural form of AI is intelligence that can sit near the flow of life or work, understand enough of what is happening, and surface itself when it is actually useful.
Most meaningful decisions are connected to what came before them. An intelligent system should be able to carry forward relevant context, understand how a situation has evolved, and avoid treating every interaction like a brand-new relationship.
More memory, more data, and more learning do not automatically create better intelligence. What matters is becoming more discerning: knowing which information still matters, what has changed, what should be ignored, and how previous experience should alter present judgment.
An answer cannot be separated from the situation around it. People, history, timing, relationships, environment, goals, and consequences all change what the right response is. I think useful AI has to understand more of that surrounding reality.
Being capable of acting does not mean acting is always right. Sometimes intelligence should speak. Sometimes it should notice quietly. Sometimes it should wait. Sometimes it should recognize that a person knows something the system does not.
I want technology to reduce the amount of life spent operating technology. The better it understands context and continuity, the less people should have to manage it, repeat themselves to it, or constantly translate their world into prompts.
I do not see intelligence as a race to remove the person. Humans carry values, responsibility, intuition, relationships, lived experience, and forms of judgment that matter. Better AI should expand what people can notice and decide, not flatten those things away.
The direction that interests me most is AI that feels less like software waiting to be operated and more like a form of intelligence with memory, context, restraint, and enough continuity to become wiser about the world it is part of.