Distributed systems, Testing and Developer tooling
Systems become easier to build when their behavior is observable, their communication is explicit, and their tests easy to reason about.
Current project
A cross-platform app for running multiple AI agent workspaces locally.
Each workspace gets isolated processes, ports, Docker services, logs, and browser state.
Agents control workspaces through MCP. Humans review them from one desktop app.
Agent-Up is open source and built for parallel web app development.
A deterministic testing solution for distributed systems. First class support for async & enumerated execution, compartmentalization and adapters.
A team-built room booking system with GitHub OAuth authentication, onion architecture and external API integration through sync and async flows.
HiveShard is a distributed runtime for deterministic 2D simulations using spatial partitioning and lockstep sync, designed to scale regardless what world size.
Aug 3, 2026AI agents should not just write code.Why web-app agents need isolated browser sessions as part of the same workspace runtime that owns source, processes, ports, logs, screenshots, and review state.
Jul 29, 2026Why use Agent-Up?Why Agent-Up is worth adopting only when parallel coding agents make runtime state, browser sessions, and human-controlled commit review difficult to keep trustworthy.
Jul 28, 2026AI-generated changes are easy to create and hard to review.How Agent-Up commits gives agent-assisted development a local commit proposal queue so logical slices can be preserved, reviewed, tested, and committed under human control.Engineering principles
I value software that is understandable, testable, and practical to evolve. The strongest systems are usually not the most complicated ones, but the ones whose trade-offs are explicit.
Good engineering starts by understanding constraints, risks, and the people who have to operate the result.
Code, tests, and architecture should make intent visible. Clever solutions are only useful when they stay understandable and maintainable.
The best tests explain what the system should do, not how it does it. But tests only make regressions easy to diagnose if the test net is fine enough.
Repeatable work belongs in tooling, pipelines, and defaults so attention can stay on design, architecture and behaviour.
Abstractions should remove real complexity without hiding the behavior developers need to reason about.
Software should be shaped so future changes are smaller, safer, and easier to review than the first implementation.
Background