Dexter Horthy, known as Dex, is the founder and CEO of HumanLayer. He writes about making AI agents reliable in production and about managing the limited working context of coding models. His work connects agent design to the engineering team’s need to understand and review the resulting changes.[1][2][3]
Start with Advanced Context Engineering for Coding Agents →
Links
HumanLayer
X LinkedIn Founder and CEO · HumanLayer
Career
Horthy says he began programming in high school at NASA’s Jet Propulsion Laboratory. Before HumanLayer, he spent seven years at Replicated across engineering, solutions engineering, product, and executive work, helping software vendors deliver applications into customers’ own Kubernetes environments. HumanLayer participated in Y Combinator’s Fall 2024 batch. Its current product brings coding-agent sessions, plans, and diffs into a shared workspace for engineering teams.[1]
Contributions
In Advanced Context Engineering for Coding Agents, Horthy describes compressing information passed between research, planning, and implementation. A model investigates a codebase, then a person reviews its findings and plan before implementation. The objective is to preserve useful decisions while removing conversation that obscures them. His examples describe his team and collaborations, rather than a controlled productivity experiment.[2]
Ideas
The 12-Factor Agents project gathers practical design principles for production software, including control over prompts, context, and execution. It identifies concrete interventions when a capable model behaves unreliably.[3]
Tools, writing & talks
Writing, research & talks
Start here
Advanced Context Engineering for Coding Agentshumanlayer.devAn account of research, planning, implementation, and deliberate context compaction in complex repositories.
Compare its research-plan-implement workflow with one task in an existing repository. Basic coding-agent experience helps you recognize the context and review problems.
Author[2]A panel of heavy AI users reports what works after OpenAI DevDay: agent loops and forks, subagents, context management, model speed and quality, local models, prompt injection risk, payments, and delegation.
Catalog workOriginal sourceHumanLayer's Dex Horthy, known for the context engineering framing, discusses why AI agents underperform and how to structure context, in an informal cooking-show format.
Catalog workOriginal source- 12-Factor Agents
github.comA practical guide to the engineering decisions behind production agent applications.