FAQ
Frequently asked questions
The questions that come up most often, answered directly. If yours is not here, reach out.
What is agentic engineering?
Agentic engineering is the discipline of building software development workflows where AI agents complete substantial tasks autonomously: reading a ticket, writing the code, running the tests, and creating a CI-green pull request, without a human steering every step. It is not AI-assisted coding (where you write and AI helps) and not prompt engineering (writing clever inputs for a chat interface). It is a structured practice with its own tools, patterns, and quality discipline. The full definition is on the agentic engineering guide.
Do AI agents really work for production code?
Yes, with the right setup. I have been running autonomous PR pipelines on a live enterprise codebase since early 2026. The honest numbers: nine agents in sequence, under 20 minutes from ticket to CI-green PR, zero manual coding. The caveats: it took eight sessions of infrastructure to get there, and the agents require well-written context files and review gates to stay reliable. "Works" is not free. The case studies have the real numbers.
How much does coaching cost?
Coaching is €50–100/h depending on scope and frequency. 10-hour blocks are available from €450, which works out to €45/h. Sessions are 1:1, focused on your actual codebase, and include async support between meetings. Full pricing on the services page.
How much does a workshop cost?
Training workshops start at €1,500 per day. The typical format is one or two days for teams of 3–15 developers. The curriculum is customized for your tech stack, and the exercises use your own codebase where possible. You leave with working tools, agent templates, and two weeks of follow-up support included.
How much does consulting cost?
Consulting for AI transformation strategy runs €800–1,200 per day. This is for CTOs and engineering organizations looking to roll out agentic engineering across multiple teams, including readiness assessments, custom roadmaps, and train-the-trainer programs.
Do you work remotely or on-site?
Both. Coaching and consulting are remote by default: Google Meet, async communication, async code review. Workshops are usually on-site because hands-on exercises benefit from the same room. Remote workshops are possible for distributed teams. Location is not a constraint for most engagements.
Do you work in German or English?
Both. Coaching and consulting can be in German or English, your choice. The blog and primary documentation are in English because English reaches a global audience, but I am a native German speaker and work daily in a German-language company environment.
What tools do you actually use?
Claude Code is the primary tool, used as the agentic development environment. The orchestration layer is a persistent AI system I built called Cairn, which coordinates specialized agents across sessions. For CI/CD, standard GitHub Actions. For sandboxing, Docker. For context management, CLAUDE.md files and a three-tier memory architecture (STATUS.md, daily journals, distilled facts). Skills rather than custom agents for most specialized tasks. Nothing exotic, nothing locked to a single vendor.
How do we start?
Book a 30-minute call. It is free and has no obligation. We talk about your situation: what your team is doing today, what is blocking you, what a realistic result looks like. If a service makes sense, I will say so clearly. If it does not, I will say that too. No pitch, no pressure. Contact options are here.
Is my code safe? What about confidentiality?
Workshops and consulting engagements use NDAs as a default. Coaching sessions work on whatever you bring, and nothing leaves the session unless you share it. I do not store code, logs, or session content. For remote sessions, screen sharing is optional: many coaches work from descriptions and code snippets rather than live access. If your org has specific confidentiality requirements, we accommodate them.
How long until we see results?
Honest answer: the first results come quickly, the compounding takes longer. After a one-day workshop, teams leave with working agent templates and a configured CLAUDE.md. The first autonomous feature build typically happens within a week or two of the workshop. The fully autonomous pipeline, ticket to merged PR without supervision, took me eight sessions of infrastructure work before it ran. Most teams reach that in 4–8 weeks with regular coaching, faster with a consulting engagement that embeds the infrastructure work.
Does agentic engineering work for non-TypeScript or non-Kotlin stacks?
Yes. The practices: CLAUDE.md context files, three-tier memory, specialized agents, review pipelines, orchestration patterns, work across any language or framework. I have direct production experience with TypeScript, Kotlin, and Python. The agent does not need me to have experience in your stack. It needs a well-written CLAUDE.md and clear conventions. Stack-specific guidance on what to put in that file is part of every engagement.
What does an agent get wrong most often?
Three things come up repeatedly. First, production hardening: rate limiting, CORS, graceful shutdown, the things that work in development and break under real load. Agents reliably miss these unless you include them explicitly in scope. Second, real-world data edge cases: an agent tests with clean synthetic data and misses the FormData ordering bug that only shows up on mobile browsers with specific image sources. Third, over-confidence on ambiguous requirements: if the spec is unclear, the agent will resolve the ambiguity in the most plausible direction, not necessarily the right one. The fix for all three is better specification, not smarter agents.
Still have questions?
The fastest way to get a real answer is a 30-minute call. No pitch. No pressure. Just an honest conversation about your situation.
Book a free intro call30 min · Google Meet · or reach out directly