Writing

Field notes from two weeks as a forward-deployed engineer at a foreign-trade company: what actually happens when an agent takes over shipping-document work, why demos and production are different animals, and how one document workflow quietly pulls a whole company into working with agents.

Vibe coding accelerates change faster than human understanding can scale. A practical look at software entropy, harness engineering, and three constraint mechanisms that keep agent-generated systems maintainable.

Extending the four-quadrant prompting framework to the project level: goal clarity × method clarity × capability yields six cells, each with its own model choice, harness setup, and verification strategy for vibe coding.

Thoughts after reading Outshift by Cisco's Scaling Out Superintelligence: the next problem is not more agents, but shared intent, shared context, and collective innovation across people, agents, and platforms.

Notes on Martin Fowler's piece on Bayer's PRINCE system: the hard parts of context engineering, the gap in multi-agent collaboration protocols, and why reliability belongs in a continuous-iteration loop.

A hands-on critique of Databricks Genie Space and Genie Code from a real migration workflow, and what Databricks may need to build next.