Writing
The writing below reflects my thought process, opinions, and evolving perspective on engineering, leadership, technology, and the decisions that shape resilient systems and effective teams.
Enterprise agentic systems should not rely on humans for every decision. The stronger pattern is metacognitive controlled autonomy: agents that know when to act, verify, escalate, or stop.
A practical look at PoVSmith and how enterprise Java teams can move from scanner-driven vulnerability tickets to generated proof-of-vulnerability tests and remediation evidence.
A practical engineering-leader interpretation of Look Before You Leap: Autonomous Exploration for LLM Agents, explaining why reliable agents need a discovery phase before execution.
Software 3.0 is not simply AI writing code. It is a shift toward engineering systems around context, tools, memory, evals, security, and feedback loops.
OpenAI's goblin incident is a useful case study in provider-side behavior drift, reward design, and why regulated enterprises need AI control planes at the workflow boundary.
A practical banking example showing how behavior contracts, evals, contract checkers, telemetry, and rollback thresholds can control LLM behavior drift.
Why productivity gains are only the starting point — and why trust, work redesign, and talent development determine whether AI actually sticks.
A practical view of why resilience belongs in design reviews, testing strategy, canary releases, rollback readiness, dependency mapping, and team culture.
A practical guide to JSON and TOON, with examples and a clear decision framework for AI-assisted development.
A practical look at how judgment, communication, and engineering management evolve when software delivery speeds up.