Spec just enough, code for quality, verify and fail fast, iterate, succeed.
About · Master of my agents
Uwe Jugel — building verifiable software with AI agent fleets.
I build small, fast, local-first tools across my project fleet and verified commits, staying in control through YAML specs, automated checks, and file-based workflows.
Explore all projects →33 Projects[1]
11,783 Verified commits[2]
1,636 / 2,683 Tracked tickets[3]
360 Studies[4]
Don't automate the creation before you automate the verification.
Background and focus
Profile
Role
Cloud/Data Architect
Education
Doctorate (Dr.-Ing.) in Data Management and Data
Visualization
Mission
Paving roads for humans and AI to build secure,
sustainable software.
Focus now
Agentic engineering — harnesses, verification, and
local-first developer tools.
Agentic & software engineering
Process & evidence
How I Work with Coding Agents (What Agents Do vs. What I Do)
What I Do (Human Control)
I define product scope, write or approve the YAML
single source of truth in spec/, set
safety and UX boundaries such as voxi's own-voice
consent checks, design canary tests, and trigger
skills manually.
What the Agents Do (5-Phase Sprint)
Harnez runs five phases: Advisory explores the scope, Dev implements the ticket, Review checks the result, Hygiene cleans up and syncs the tracker, and Retro records what was learned.
What Fails & How Canaries Catch It
In Loom's module migration, two semantic mistakes around quit and handled results passed the test suites; review caught them. In Cati, mismatches around explicit dimensions led to a geometry planner and dedicated regression coverage.
My Take on Harnesses & Local LLMs
Problem Harnesses are getting too big and too opaque. They eat my time and hide what the agent is actually doing.
Answer Frontier models, a few generic helpers, and manually triggered skills and commands keep me in control. When a project goes spaghetti (looking at you, emojig + Zig), a frontier model can help clean it up.
I keep an eye on local LLMs too — but I won't sink real time into them until smaller models need less hand-holding.
Toolbox & Timeline
Daily drivers
Playing with
2022 · First contact
Playing in the OpenAI Playground with
code-davinci in private beta. Hooked from day one.The hand-holding era
Tuned my own tools from the start — learning to build guardrails, not just prompts.
2025 · Trust, but verify
Shifted from babysitting every step to verifying the result. Automate the checks, then let the agent run.
2027+ · Wide open
Curious where this leads. The tools change every month — and I want a front-row seat.
Written and compiled offline
Books
Written the same way the software is built — drafted with agents, verified like code, compiled offline as mdBooks. Now shelved in the library — books, decks, and PDFs included.
Agentic Software Development
The Ubunatic WayThe playbook behind this site: manifesto, canary-first development, the language "good parts", and the golden-change playbook — how one developer runs a fleet of focused apps with AI agents.
Emojig
Chronology of a Zig EndeavourSix weeks, one binary, zero daemons: the week-by-week story of building a <900 KB terminal emoji picker in Zig — real numbers, dead ends included.
The Style Codex
A Fancy-Styles Test BookThe living golden sample that keeps the book engine's theme honest — typography, callouts, tables, figures, and code, one style family per page.
Research and leadership
Background
A fuller CV from before the agentic pivot — restored from the earlier version of this site and still being written up.
Data & Architecture — Research
The doctorate asked one question: why send a million rows to a chart that is 400 pixels wide? The answer — let the visualization drive the query — became a line of work on pixel-perfect data reduction inside the database.
M4: A Visualization-Oriented Time Series Data Aggregation — PVLDB 7(10), 2014, selected among the best papers of VLDB 2014. Pixel-perfect line charts from data reduced by orders of magnitude.
VDDA: Visualization-Driven Data Aggregation in Relational Databases — The VLDB Journal 25(1), 2016, the invited extended version: data reduction operators for all common chart types, in plain relational algebra.
Dr.-Ing. thesis: Visualization-Driven Data Aggregation — full text (PDF), rethinking data acquisition for visualization systems.
Engineering Leadership & Delivery
Over a decade of hands-on technical leadership: guiding distributed engineering teams, architecting high-throughput data platforms, and mentoring engineers from early-stage prototypes through production scale.
Technical Architecture & Delivery — Leading cross-functional teams across systems programming (Go, Rust, C++), infrastructure automation, and full-stack web platforms. Emphasizing specification-driven contracts, continuous verification, and zero-defect delivery pipelines.
Mentoring & Engineering Culture — Developing engineers into autonomous system owners through pair programming, rigorous code review standards, and disciplined architectural playbooks.
Meanwhile, the record lives on Google Scholar and LinkedIn.