About Dmitry Buykin
Dmitry Buykin builds AI agent systems for shipping operations at Maersk, where his conference bio lists him as Applied AI Lead and staff software engineer. According to the same bio, his background spans applied mathematics, Nordic payments, trading systems at Deutsche Bank and machine-learning adoption at DataRobot.
He spoke twice at AI Engineer events in 2026. "Tribal Dungeons of Global Shipping" covered running agents on shipping procedures across many countries. His November talk dealt with the maintenance problem that follows: coding agents produce more change than anyone can check, and a system built by them has to be kept honest with evidence.
Career Highlights
- Applied AI Lead at Maersk, working on agents that carry out shipping operations procedures
- Speaker at AI Engineer World's Fair 2026 ("Tribal Dungeons of Global Shipping: AI Agents at Global Scale")
- Earlier work in payments, trading systems and machine-learning adoption, per his speaker bio
Notable Positions
On Prose as the Real System
By his estimate about three quarters of the system he describes is prose instructions, written by coding agents and followed by production agents. He treats that prose as something to engineer: mutation-test it, compile it, keep one copy, and have business experts sign the text they can read.
On Moving the Line Between Prose and Code
When agents keep taking the same path through the instructions and it agrees across countries, the path becomes a workflow tool and the prose is deleted. What varies stays in prose an expert can read. Experts own the decision table, the model picks a row, and a person takes over when no row fits.
On Context as a Cause of Action
Buykin found that a model allowed to load everything invents work from whatever it sees. His runtime decides what the model sees and passes large data as handles. See Context Engineering.
On Evidence
He ranks evidence by what it proves, with a replay on real tasks at the top, and requires every new check to fail once on purpose before it is trusted. A second agent with a checklist reviews the first agent's work without seeing its reasoning.
Key Quotes
- "Coding agents made code cheap. They made entropy even cheaper." (on maintaining agent-written systems)
- "Whatever sits in its context becomes a reason to act." (on limiting what the model sees)
- "Verification speed is the limit." (on where time goes)
- "Conway said the system copies the org chart. Ours copies whoever corrects it." (on whose habits the system inherits)
Related Reading
- Maersk's Dmitry Buykin: Agents Made Entropy Cheap - Full breakdown of the talk
- Context Engineering - Deciding what the model sees
- Generator-Evaluator Pattern - Why the author shouldn't judge its own work