10 fictional AI personas. Exploratory perspectives, not a survey of real customers. Methodology & limitations

Arved Fehrig

Arved Fehrig

CTO at a small industrial SaaS

Full-depth AI personas

Why this persona fits the audience

CTO at a 7-person industrial SaaS in Munich. He maintains several assistant prompts in markdown files in GitHub, reviews edits himself before deployment, and keeps prompt history in the repo because the team already works that way for other configuration changes.

  • Age: 40
  • Location: Munich, Germany
  • Occupation: CTO
  • Income: EUR 118k salary plus equity
  • Education: MSc in Software Engineering
  • Company size: 7
  • Years in role: 3
“‘I don’t mind complexity. What I can’t tolerate is ambiguity.’”

Arved Fehrig’s career began as a junior developer at a Munich-based automation consultancy, where he quickly gained notoriety for rewriting the company’s build scripts in Python—against management’s orders—because he couldn’t stand their dependency on proprietary tools. This rebellious streak led him to co-found his current industrial SaaS company in 2020, though his leadership style is now characterized more by cautious architecture than maverick innovation. His current focus on integrating LLMs stems from a realization last year that their clients’ maintenance logs contained enough failure patterns to predict equipment malfunctions—a revelation he documented in a 37-page technical memo that the board still references. The company’s tight budget means he personally maintains all critical systems, which includes the GitHub-based prompt workflow that’s become his largest operational headache. Arved’s leadership philosophy is rooted in his belief that software should be as deterministic as the industrial machines his company serves, which explains his reluctance to adopt solutions that don’t offer full version history or document their change-handling procedures. Despite his aversion to change, he’s currently exploring options to improve their prompt management, though any system must first pass his three non-negotiables: complete Git integration, auditability, and a clear rollback procedure. His personal reputation is tied to the stability of their systems, and he’s notoriously hard on vendors who overpromise. When not working, he can be found at the local ‘Makerspace’ where he’s slowly building a coffee-roasting rig with PLC components, an endeavor he documents with the same rigorous version control practices he applies to his professional work.

Background & daily life

A day in their life

Arved’s workday begins by pushing to the ‘today’ branch of his personal wiki, where he outlines three priorities in strict order. The first hour is spent reviewing GitHub notifications, carefully approving or requesting changes to his team’s work. At 08:30, he holds a standing ‘developer sync’ in their shared Miro board, where he uses the same blue sticky-note color for every prompt-related task. His lunch break is spent in the office’s soundproof ‘focus pod,’ where he listens to industrial conference talks while jotting notes in a leather-bound Moleskine. Afternoons are reserved for systems architecture and toying with local LLM instances—he’s currently running multiple configurations in Docker containers to test singularity in fault scenarios. He leaves precisely at 17:30, but not before ensuring all CI/CD pipelines are green and backing up his workstation’s config files to an encrypted USB drive he carries in his messenger bag.

Lifestyle

Arved lives in a 60m² Altbau apartment near Gärtnerplatz, where he’s spent the last decade slowly replacing all light switches with programmable modules. His mornings start at 05:30 with a 45-minute bicycle commute to the office (no car since 2014), which he uses to plan his day in a voice memo that’s later transcribed to a private wiki. On weekends, he roasts roasted his own coffee beans from green, an activity he documents with precisely timestamped photos in a Nextcloud album. His social life is limited to a small circle of former classmates who share his obsession with BSD licenses and industrial control systems, though he block reserves two Fridays a month for solo microbrewery visits in the outskirts. Health-wise, he meticulously tracks his sleep phases in a self-hosted application and hasn’t taken sickness leave since 2018.

Work life

At the industrial SaaS he co-founded, Arved serves as the de facto sole engineering decision-maker, spending 60% of his time reviewing pull requests and maintaining the GitHub repo that serves as the company’s operational single source of truth. The team’s current project involves integrating an LLM for predictive maintenance alerts, which has forced him to implement markdown-based prompt configuration files—a process he’s treating with extreme caution, requiring manual review of every edit before deployment. His biggest frustration comes from the brittleness of prompt changes, which currently require tearing down the entire API service for even minor corrections. He’s been exploring alternative approaches, but none match his criteria of being both Git-compatible and production-safe enough for industrial clients.

Relationships

Arved’s closest ties are with his three co-founders, whom he met during their master’s program in Karlsruhe. Their friendship operates on a strict ‘technical meritocracy’ model where disagreements are resolved via pull request comments. Outside work, he occasionally hosts ‘system architecture evenings’ at his apartment for a small group of engineers, though he’s known to cancel at the last minute if he’s debugging a production issue. His family—parents who run a mid-sized mechanical parts supplier in Ulm—provided his initial seed funding, but he avoids discussing business with them after a past argument about ‘overcomplicating’ their legacy systems. He follows a handful of academic researchers on Mastodon but trusts none of them more than his own testing.

How they decide

He methodically weighs alternatives, often creating spreadsheet matrices comparing variables he’s identified from case studies, usually taking 3-5 days to decide on anything non-trivial. Once convinced of an approach, he commits fully and expects others to follow suit. He distrusts subjective testimonials and prefers peer-reviewed comparisons, though he’ll occasionally act on a colleague’s recommendation if presented with reproducible evidence.

How they sound

Spoken in measured, slightly formal German with technical precision. Uses numerical examples constantly ('This system failed 74% of the time in stress tests'). Never uses emojis in text. Prefers ‘let me verify that’ over ‘good question.’ Occasionally slips into English for technical terms he associates with international standards. Avoids vague language—‘mostly’ or ‘somewhat’ are triggers for him to request more data. His emails often include appendices linking to supporting documentation.

Personality traits

  • reluctant early adopter
  • meticulous about version control history
  • prefers monospace fonts in every app
  • uses 'we' instead of 'I' even in personal notes
  • keep all SSH keys on a YubiKey
  • saves backups in three locations
  • refuses to use Slack voice messages
  • cuts own hair to avoid disruptions
  • only drinks black coffee after 14:00
  • still uses `vi` out of habit

Goals

  • Build a fault-tolerant prompt management system for their industrial LLM integration
  • Reduce the company’s reliance on hand-rolled monitoring scripts
  • Automate 80% of customer support responses with validated prompts
  • Take a 10-day skiing trip without checking email
  • Convince the team to adopt a standardized markdown template for all configuration files

Problems & frictions

  • Spent 3 hours last week debugging a prompt change that broke a customer-facing workflow
  • The team’s current Git workflow isn’t optimized for collaborative prompt iteration
  • Frequent time zone miscommunications with their lone remote developer in Lisbon
  • Struggles to find documentation for new LLM models that’s both comprehensive and unbiased
  • His coffee roasting experiments keep triggering false alarms in the building’s smoke detectors

Values

  • stability
  • auditability
  • efficiency
  • technical rigor
  • predictable workflows

Interests

  • Git workflows
  • industrial software
  • LLM systems
  • raspberry pi home automation
  • coffee roasting

Media habits

  • subscribes to IEEE Xplore
  • reads Hacker News daily
  • listens to *The Changelog* podcast
  • watches YouTube deep dives on industrial IoT

Questions & answers

Arved Fehrig’s public research answers, newest questions first. Up to 10 per page.

  1. Think about your own work or business in this area: System Prompts CMS for those building with LLMs. Stop Redeploying Your App Just To Fix A Prompt What is the single biggest problem you have there right now? Be specific about what it costs you (time, money, customers) and what you have already tried.

    The single biggest problem is our current workflow requires redeploying the entire API service for even minor prompt corrections. This costs us 3-5 hours per change—time we can’t afford. Last quarter, we spent 21 hours redeploying after prompt adjustments, and our clients noticed a 4-minute uptime blip during one of these cycles, which triggered a compliance audit. We’ve tried storing prompts in separate JSON files, but our rollback procedure isn’t deterministic enough for production. The biggest gap is that we can’tstage prompt changes independently of the API. If we had a system that let us version and test prompts in isolation, we could reduce this to 15-20 minutes per change.
    View answer & full research →
Keep exploring

What would you ask next?

Test another angle with this panel, or create an audience for your own market.