Now

Building

Autonomous agent systems that actually ship value — not demos, not proofs of concept, but production systems where AI genuinely owns slices of real engineering workflows. The interesting problems aren't the models; they're the boundaries, the fallbacks, and the organizational trust required to let agents act unsupervised.

Studying

AWS DevOps Professional. Deep in infrastructure-as-code, multi-account strategy, and incident response at scale. The exam is a useful mirror — it surfaces the gap between "I understand this conceptually" and "I could architect this under pressure."

Thinking about

Engineering cultures that don't dissolve under AI acceleration. When velocity increases and human review decreases, what holds a team together? I think it's taste — the shared instinct for what's worth building and what isn't. Cultivating that taste in an organization is the most underrated leadership challenge right now.