Funkeltronic
DE

A founding team in Germany, 2026

Vertical AI agents with a harness.

We build custom agents for skilled work in regulated industries. Each one comes with the scenarios, evaluations, traces and guards a reviewer needs before signing off.

Talk to us How we build

Yes, that is a cat on a unicorn. The saddle is the point.

Cartoon cat riding a white unicorn with a rainbow mane

What we build

Three things, and they belong together. An agent for one job, an architecture sized to that job, and the harness that makes the agent testable.

Vertical agents

An agent for one job in one industry. It knows the forms, the norms and the exceptions, and it stops where a human has to decide.

Custom architectures

A single prompt rarely carries a whole workflow. We split the work into an orchestrator, specialist agents, tools and an audit log, and we keep the number of parts as small as the job allows.

Harness engineering

The scaffolding around the agent: scenarios from real cases, evaluation runs, traces, guards and replay. It is the part that turns a demo into something a team can rely on.

Architecture

The point where a human signs is a design decision we make first. Everything in front of it is built so that the signature means something.

Agent architecture with five stages A request enters, an orchestrator plans, specialist agents work with tools, the harness checks the result, and a human signs. Harness: scenarios, evaluations, traces, guards, replay guards traces evals replay permissions Request form, drawing, mail Orchestrator plans, routes, stops at open questions Rules lookup Calculation Document draft tools and sources Human reviews, signs audit log
1 of 5

Harness engineering

A harness turns an agent into something you can test. We write scenarios from real cases, run the agent against them, keep every trace, score the result against a rubric and fix what fails. The gate stays closed until the score holds.

Scenarios
Real cases with a known good outcome, including the ugly ones.
Evaluations
Automated runs that score the agent on every scenario after every change.
Traces
A record of each step, tool call and source, kept for review and replay.
Guards
Hard limits on scope, data and actions that a prompt cannot override.
Replay
Run a past case again on a new version and compare the two traces.
Permissions
Which tools the agent may use, on which data, and when it has to ask.
The harness loop Scenarios feed a run, the run produces traces, traces are scored, failures are fixed, and the loop starts again. A release gate opens only from a passing score. Scenarios Run Score Fix Trace every change, every scenario release gate opens on a passing score
Scenarios, run, trace, score, fix. The gate opens on a passing score.
The cat on the unicorn, annotated Labels point to the horn (the ambition), the cat (checks whether the saddle holds), the saddle (the evaluations), the reins (permissions) and the tail (the sparkle we keep). cat: checks whether the saddle holds horn: the ambition reins: permissions saddle: the evaluations tail: the sparkle we keep

Why the cat

The unicorn is the goal every founding team is told to chase. The cat is the one who checks whether the saddle holds before the ride.

We named the company after the sparkle ("funkeln" in German) and gave the cat the job of checking the harness. In practice that means the fun part, an agent that drafts a permit in minutes, comes after the boring part, the scenarios it has to pass first.

The logo is an AI-generated draft. The cat is staying.

Fields

The method stays the same, the field is still open. Six fields on the table. None is decided. We score them this autumn and take one or two into customer interviews.

01

Security

Agents that test and defend AI systems, and the compliance work that EU product rules create for makers of connected devices and software.

exploring
02

Education

AI literacy for companies, and for people who have never used an assistant. Training first, a product later if the demand is there.

exploring
03

Care

Assistants for carers, and products against loneliness for older people. Who pays is the open question.

exploring
04

Commerce

Agents that buy on behalf of people, and shops that are ready to be found and read by agents.

exploring
05

Personal products

Cards, maps and small apps generated for one person or one occasion, and thrown away after use.

exploring
06

Infrastructure and administration

Agents for planning, permits and public services, where staff are short, the work follows forms, and a human signs at the end. This is where we expect the harness to matter most.

first interviews

How we decide

Each founder scores each field on six criteria, alone and before the meeting. Weighted scores decide which one or two go into interviews. The weights are a proposal until the team has changed them once.

  1. 1 Desk research
  2. 2 First filter
  3. 3 Interviews
  4. 4 Decision
Criteria and weights (maximum weighted score 65)
  • Competence3
  • Customers3
  • Resources2
  • Excitement2
  • Market2
  • Timing1

Team

Five founders. Profiles follow once each of us has written one, which is slower than building a website.

  • C
  • S
  • M
  • F
  • N

Between us: software architecture, AI enablement in large organisations, product work, and contacts in energy, utilities and local government. Roles for the evaluation phase: idea owner, market research, customer access, prototyping, and process.

Talk to us

If you run a team where the work follows rules and forms and someone has to sign at the end, we would like to hear what slows you down. Thirty minutes, no slides.

hello@funkeltronic.ai