I’m a software engineer based in Tallinn. My work has taken me through banking, payments and systems where the details matter.
These days I work on consumer payments at Nelo. Outside that, I’m building personal tools and experimenting with how I work with coding agents.
Some of the places along the way
Nelo · ANZ Plus · BAE Systems Applied Intelligence
02 / workspace.app
My custom herdr setup
The conversation is only part of the picture.
I want the plan, the decisions and the state of the work in view. So I customise herdr around the way I use Claude: a workspace tree, the conversation, and a task mod alongside it.
The interesting part
What happens when it doesn’t go to plan?
Pick a stage, or play the walkthrough.
▣herdr / my workspaceInteractive illustration · generic task
plannerkeyboard-navigation+
An example conversation
Luke
Let’s make the project list work with a keyboard.
Luke
Keep focus visible, use the arrow keys to move between items, and leave normal Tab navigation intact.
↳
The plan and acceptance criteria live with the task.
❯/plan keyboard-navigation
Example workspaceSkills + mods + workflowsNothing here runs a real task.
↔ Scroll the workspace to explore all three panes.
02 / 07
Put the decisions somewhere they can stay.
I work through the design and acceptance criteria with Claude. A custom mod keeps the plan visible beside the conversation.
Skills / the instructions
I keep repeatable ways of working in skills, so the context and expectations don’t need rebuilding in every conversation. Planning, task work, reviews and pull-request updates each have a place.
/plan/task/review/pr-update
A simplified, interactive example of my workflow. Generic commands; no real task is running.
Under the hoodA short tour of a much bigger system.
+
This page shows one simplified task. My actual setup connects an ongoing plan to individual task files, agent roles, decision gates and a mod that keeps the current state beside the conversation.
01 / before the task
A plan that stays alive.
The plan holds proposed work, decisions, questions, dependencies and blockers. I accept, defer or decline tasks. A launched task gets its own tab and keeps its connection to the plan.
02 / before building
Three sources. Then my call.
Separate agents validate the tickets, documentation and current code. They record supporting evidence, concerns and contradictions. I approve validation, then work through the goal, criteria, subtasks and planned tests before approving implementation.
03 / while it runs
State that survives a session.
Goals, test evidence, review findings, worktrees, PRs and follow-ups live in structured task state. The mod displays it beside the chat. Agents record their results there; the workflow checks the gate before advancing.
04 / before it closes
Evidence all the way through.
Every criterion needs evidence and the regression run must pass. Claude and Codex independently review the same commits. Findings need a resolution. PR checks, further review and ticket updates remain part of the task; it closes after merge.
Generic names here: /plan/task/reviewMy judgement stays in the loop.
03 / things-i’m-making
A few open projects.
Small tools with plenty going on underneath. Open one for a closer look.
From this desk to your next project
Let’s make something good.
A product to build, a system to untangle, or a team doing interesting work. I’d like to hear about it.