Building where the work is hardest.

Legacy modernization

Most of the meaningful technology work in the world today still happens inside organizations that have decades of accumulated systems — bank cores, insurance policy engines, public-sector registries — running code that predates the engineers maintaining it. The temptation is to rip and replace. The reality is that these systems carry institutional memory, regulatory commitments, and risk surfaces you cannot just port.

Good modernization work respects what is there. It maps the seams, retires functionality at the rate the business can absorb, and resists the engineer's instinct to start from a blank page.

Responsible AI in enterprise

The interesting question for AI in enterprise is rarely the model. It is what the model is allowed to touch, what it is required to log, who is accountable when it misfires, and how a regulated organization tells a coherent story about it to its board, its auditors, and its customers.

I am wary of two failure modes here too: the one that treats AI as a magic layer that obviates governance, and the one that treats it as a threat to be sandboxed into uselessness. Neither serves the people doing the actual work.

The Global South question

My longer-term hope is to see the dividends of the AI revolution reach the Global South — not as passive consumers of imported tools, but as builders, freelancers, and entrepreneurs with genuine access to the digital economy.

That requires more than market access. It requires data, compute, and participation in the upstream decisions about what the systems are optimized for. If the next wave of frontier models is trained, fine-tuned, and red-teamed without us, we will inherit their assumptions.