Qualitative
The journal captures what shows up in the work: good prompts, broken assumptions, clean handoffs, and needless loops.
Organization theory meets agentic coding
I get your team moving. Then you take it from there.
A workshop, the initial setup, and clear next tasks get you through phase 0 and 1. After that, your team carries phase 2 and 3 forward on its own.
Where teams get stuck
Agentic coding rarely fails in the first demo. It fails in day-to-day work when review, quality gates, and learning do not grow as a system.
Review becomes the bottleneck
Your fastest people get stuck in review while agents produce more and more. That cuts throughput and makes quality depend on a few calendars.
Quality gates keep shifting
What gets checked before merge changes from person to person, repo to repo, or day to day. The agent never sees a stable target.
The learning loop is missing
Without a journal and measurement, every good or bad outcome stays a one-off. The team repeats mistakes instead of turning them into better prompts, rules, and tasks.
The approach
I combine organization theory with agentic coding. We first build the working shape in which agents can contribute reliably: roles, gates, handoffs, and clear ownership.
I track two levels with you: qualitatively in the journal, quantitatively in the signals from hooks, logs, SQL, and traces.
The phases
I get you through phase 0 and 1 hands-on. From phase 2 on, you have the concepts and the concrete tasks to keep going on your own.
- 0
Foundation
Branching, review, CI, and clear handoffs are in place. From here on, agents work inside a frame that makes mistakes visible early.
- 1
Guided rollout
We set up the first building blocks together and test them on real work. The team understands why the rules are shaped the way they are.
- 2
Team scales itself
You add more subagent prompts, skills, AGENTS.md facts, and reviews yourselves. Examples turn into routines.
- 3
Controlled flow
Agents carry most of the flow, and people step in at the points that matter. The learning loop keeps quality and speed aligned.
Day 1
Day 1 is workshop and setup in one move. I put the foundation in place with you and work through the concepts until it is clear how you continue without me.
- Phase 0 and the first parts of phase 1 are anchored in day-to-day work.
- The team understands the core ideas behind review, roles, handoffs, and the learning loop.
- You leave with a prioritized task package for the next steps.
In the best case, one shared start is enough. After that your team implements the rest on its own.
Four building blocks are enough
You do not need a huge library to get started. These four artifacts carry most of what later becomes reusable.
subagent prompts
They cut work into units with a clear scope, clear inputs, and a clean output for a subagent.
skills
They capture recurring workflows so good steps do not have to be reinvented every time.
AGENTS.md
This is where facts, rules, and local conventions live. Context does not have to be restated in every prompt.
handoff contract
It defines what has to be delivered when work moves between agent and person or between two agents.
The learning loop
Agentic coding gets better when you feed outcomes back into the setup on purpose. The learning loop turns that into a routine.
Quantitative
On top of that, you track measurable signals from hooks, logs, SQL, and traces. That shows where agents save time, get stuck, or lose quality.
Specialization
My background is data engineering. That is where weak handoffs, shallow reviews, and fragile pipelines get expensive early. The patterns behind it apply in any stack.
Eike Stürmer
About me
I am Eike. I work where ways of working and code meet. My goal is a setup that keeps moving without outside help after a short period of shared work.
Let us look at where you stand
If you are stuck between a first experiment and a reliable flow, we can talk about phase 0, day 1, and the next concrete tasks. Booking is the fastest path, email works too.