Small Team.
Unreasonable Output.
veith.dev builds your software, from internal tools to whole products, and sticks around to run it. We work with AI agents all day. That's why the output doesn't match the headcount.
Senior engineers · Stuttgart, Germany · We publish how we work
Software Economics Changed. Most Teams Didn't.
AI agents can now do the volume work of software development: writing, testing, refactoring, reviewing. But tools don't change output, workflows do. Companies bought licenses and kept the old process; agencies kept billing hours the old way. That gap is where your budget goes.
Licenses ≠ Leverage
Companies buy AI tools and see single-digit gains. Without context and review gates, an agent is a fast typist with amnesia.
Agencies Bill the Old Way
Layers of project management sit between you and the person typing. You pay for that coordination whether anything ships or not.
Hiring Doesn't Scale Down
A real dev team means several salaries and months of ramp up. Most SMB software doesn't justify that. It just needs to get built.
Slop Is the Real Risk
Unreviewed AI code is a liability. That's a review problem, not an AI problem. So we built the review in: hard gates, senior sign-off.
How Working With Us Runs
The result: Output like a big team, a bill like a small one, and someone who actually owns the quality.
How We Ship: The Agentic Delivery System
No slides about AI here. This is the system we actually work in, every day. We publish what we learn from it.
Project Context
Every codebase carries a rulebook the agents read at the start of each session: conventions, architecture decisions, what 'done' means here. Nothing gets explained twice.
$ cat CLAUDE.mdStack: Next.js 16, TypeScript strict, Tailwind v4Every change ships with a verified buildUI copy: direct, two-beat, no filler# read by every agent, at the start of every session
Skills & Workflows
Recurring work is codified into reusable skills, from review checklists to release routines, so quality doesn't depend on someone having a good day.
Connected Tools (MCP/CLI)
Agents read and act on the real project surface through MCP servers and plain CLIs: the repo, the issue tracker, CI, the browser, the database. No copy-paste middleman.
Parallel Agents
Independent tasks fan out to several agents at once. One refactors while another writes tests. Throughput scales with compute, not headcount.
Senior Review Gates
Every line lands behind deterministic gates: automated checks, hooks, and a senior engineer who owns the result. Agents do the volume. Engineers sign the work.
What We Do
Four ways to work with us. Each one starts as a fixed package.
Custom Software Development
Development Partnership
Team Enablement
Vibe-Code to Production
Know What You Get. Before You Sign.
Four fixed packages, each a scoped way into a longer engagement. Fixed days, defined contents, a named audience. Fixed price, agreed before we start. No hourly meter.
Secure Delivery Enablement
10 days- For
- Development teams of 3 to 8 engineers already shipping AI-assisted code, with no pipeline or review gate between them and production.
- Scope
- One team, one primary codebase, one CI platform.
$ cat enablement-10d.mdday 01 threat model, secrets audit, access reviewday 02-03 ci/cd: build, test, staging, one-step rollbackday 04-05 security gates: sast, dependency + secret scanningday 06-07 agent workflow wired inside the gatesday 08-09 workshop, your codebase, your whole teamday 10 handover: runbook, named owner, 30-day follow-up# every artifact lands in your repo, not in a slide deck
Leads into Team Enablement
Ask about this package →Production Audit
5 days- For
- Teams running an AI-built or fast-grown app that real users already depend on, with no test suite and no clear picture of what breaks next.
- Scope
- One repository, one environment, read access to what runs today. Anything larger we requote before we start.
Day plan
- day 01-02
- read: security, data model, dependencies, secrets
- day 03
- infrastructure, hosting, gdpr posture
- day 04
- findings ranked, each with a fix and an estimate
- day 05
- written report, then the 90-minute walkthrough
Scoping Sprint
5 days- For
- Founders and teams with a project they can describe but not specify, who need a real number before they can release budget.
- Scope
- One product, one team. The thin slice runs in your stack, on your infrastructure.
Day plan
- day 01-02
- requirements workshop, written specification
- day 03
- architecture and stack decision, reasons recorded
- day 04
- one thin end-to-end slice running in your stack
- day 05
- build plan in increments, fixed-price quote valid 60 days
Leads into Custom Software Development or Development Partnership
Ask about this package →Agentic Kickstart
2 days + 4 weeks- For
- Senior developers who can already build anything, and for whom AI tooling still feels slower than doing it themselves.
- Scope
- One developer, one primary codebase, one to one. Not a classroom course.
Day plan
- day 01
- your repo, your tickets: where delegation pays and where it does not
- day 02
- context, skills, tool access: the rulebook agents read
- week 01-04
- one paired session a week, on real work
- end of week 04
- your workflow runs without us
Leads into Team Enablement
Ask about this package →Who's behind this
Engineers, Not Slides.
veith.dev is senior engineers in Stuttgart. We also build and run 2kw.ai, an AI platform working in industrial production. And we practice what we sell: agents built this site under our review, and we write up how.
Questions, Answered
What does veith.dev do?
veith.dev is a software development company in Stuttgart, Germany. We build custom software (internal tools, integrations, web applications, MVPs) and run long-term development partnerships for companies that need an engineering team without hiring one. We work AI-native: agents do the volume, senior engineers review and own every result.
What does AI-native development mean?
AI-native means the workflow is built around AI agents rather than just assisted by them: project context files the agents read on every session, codified skills and workflows, tools connected over MCP and CLIs, automated review gates. Agents do the volume work. Senior engineers set the standards and own the result. That's how a small team ships like a large one.
Who do you work with?
Founders and small to mid-sized companies in the German-speaking region and the EU that need software built and kept running, plus development teams that want to adopt agentic workflows themselves.
Can you make a vibe-coded app production-ready?
Yes. If you built an app with AI tools like Lovable, Bolt, or Cursor and real users now depend on it, we make it production-grade without a rewrite: test suite, CI/CD pipeline, security and GDPR hardening, and a senior engineer who owns the result. You keep building with AI; we install the engineering system around it.
Why is a small AI-native team faster than an agency?
The economics changed. An agency bills people-hours across layers of coordination; we run agents in parallel under senior review, so most of that overhead disappears and throughput per person multiplies. You pay for shipped software, not seat time.
Do you work remotely?
Yes. We're based in Stuttgart and work with clients across the German-speaking region and the EU: on-site where it helps, remote by default.
Do you work with fixed prices?
Yes. Every engagement starts as a package: a fixed number of days, defined contents, and a price agreed before we start. You carry no estimation risk and there is no hourly meter. Longer work that follows a package is scoped and quoted from what the package found.