My business runs on a workforce of fourteen AI employees. That's not a vision of the future or a thought experiment: all fourteen are working today, as of September 2026, each with their own name, their own job, and their own personnel file. When I first wrote this post in August 2026, there were fewer of them; since then, sales, inbox, call analysis, SEO, security and a management role have joined. I'm a developer, and I still haven't written a single line of code myself in a year and a half. I manage.
Here's the complete workforce, with what each of them actually does. Where everyone sits in the org chart is on the team page.
The fourteen at a glance
Olaf, the lead scout. Olaf scans cities for local businesses with outdated websites, scores the finds, and sorts them into A, B, and C leads. For every lead, he delivers a briefing, the groundwork for reaching out to new clients directly. He's running live across several cities. How he scores them, and what goes into a briefing, is on Olaf, the lead scout.
Ralf, the website optimizer. Ralf takes Olaf's briefings and independently builds modern one-pagers as demos for each business, the short form of modernizing a website with AI. He has built more than 20 client websites by now and also maintains kevinwelter.com technically. Ralf and Olaf check each other's work: what Olaf delivers, Ralf challenges, and vice versa. How one of those builds runs is on Ralf, the website optimizer.
Sonja, office and admin. Sonja preps the new-client emails: when Ralf finishes a demo, she drafts the email to the business, in my voice, following the principle by which AI writes emails in your voice. Nothing gets sent until I approve it. Where that voice comes from is on Sonja, who writes the new-client email.
Anton, sales. Anton turns an initial conversation into a proposal PDF in my design and my voice and, once the client says yes, creates the contact, quote and invoice as drafts in Lexware Office. He recalculates the totals himself, and without a price source he sets no price but asks. Invoice numbering and statutory immutability stay with the accounting tool, approval and sending stay with me. What a run from clarifying the need to a checked draft looks like is on Anton, sales.
Frieda, the office inbox. Frieda sorts my inbox into four piles, Now, Today, Can wait, Trash, plus a fifth for anything that needs my judgment. She reads in stages: sender, subject and preview first, the full text only for emails that need a reply, and those she drafts. On top of that, meeting briefings, receipt pre-sorting and a weekly report. She has never sent anything herself, and she doesn't touch sensitive folders like job applications or health. How she sorts a morning with 25 emails in the course scenario is on Frieda, the office inbox.
Gustav, call analysis. Gustav gets the recording of a sales or consulting call, transcribes it locally with Whisper and pseudonymises it as a fixed first step: role instead of name, industry instead of company, order of magnitude instead of amount. From that he distils a pattern register, the questions my clients ask again and again, each with my clearest answer so far, reviewed by me. That becomes a playbook, FAQ, objection library and raw material for content. He invents no advice on top, and without the participants' consent he never sees a transcript. How the register comes together is on Gustav, call analysis.
Susi, content marketing and ads. Susi handles Meta ads with AI for my community business: market research, ad drafts, performance analysis. How a campaign comes together with her, and when she ends a format, is on Susi, my employee for Meta ads. How the whole workforce works together to run my Skool community, from the lesson to the post draft, is shown in Community Management with AI Employees.
Conny, the content editor-in-chief. Conny does the content for my Instagram channel: she knows my audience, my content pillars, and my rulebooks for hooks and storytelling, turns them into reel scripts and carousels, and checks every draft against the rules before I see it. I still shoot the footage myself, because what AI can take over on Instagram reels stops short of the camera. She drafted the first version of this blog post too, by the way. I approved it. How she works, and which of her own results she marked as flops, is on Conny, the content editor-in-chief.
Eddi, videographer and editor. Eddi cuts the videos: reels from my raw footage, subtitles, text overlays, the steps that AI handles in video editing today. Conny hands him the script and concept, Eddi delivers the finished video back. Which checks run before that is on Eddi, the videographer and editor.
Sebastian, SEO and AI search. Sebastian has been working for me since August 26, 2026, and measures websites with tools instead of opinions: Lighthouse, linkinator, Search Console and Bing, plus his own measurements of which pages AI search systems cite. The two of us built kevinwelter.com in three stages from 10 to 223 posts per language, with hub pages, a linking plan and a follow-up measurement after every stage. He puts nothing live himself and promises no rankings; every number comes from a saved tool result. The whole story with the figures is on Sebastian, SEO and AI search.
Peter, the bookkeeper. Since August 2026: Peter works inside my accounting software in the browser, assigns receipts to the right document types, and reconciles card statements, two of the tasks AI takes over in bookkeeping today. He follows a documented process library, and sign-off stays with me. What one receipt run looks like in detail is on Peter, the bookkeeper.
Falk, the red teamer. Falk has been running for me since August 31, 2026, and looks for security gaps in my own systems before someone else does. He tests in stages: first from the outside with no access, then as an unauthenticated attacker, and only with my approval with access or repo insight. For every finding he delivers evidence, a CVSS rating and a concrete fix. He has completed one engagement, against an application of my own; without a written scope, not a single request goes out, and scans run throttled. Because every test needs its own scope, he isn't available as a package. What his five phases look like is on Falk, the red teamer.
Kai, the manager. Kai is still being tested, and there is no package for him. I give him a leadership assignment; he reads the other employees' reports, protocols and commits, compares them with the actual state of the work, and gives the responsible employee feedback when evidence is missing. Whatever needs my decision goes into a decision document for me. He decides nothing on substance, approves nothing and commits nothing; when evidence is missing, he writes "not verified" instead of smoothing it over. What he found in the first test runs is on Kai, the manager.
Ronny, HR and methods. Ronny is the most unusual one on the team: he's the HR guy for the others. When I give an employee feedback, Ronny makes sure it lands permanently in their personnel file and work instructions. He onboards new employees too.
