HumanITy Logo
ENDE

My AI employee

Gustav hears every conversation once, yet remembers every recurring question, never the person behind it.

Gustav runs in my own business and is available at the same time as a ready-made AI employee package in the Claude Practitioners community, so you can hire him into your own business too.

I hand Gustav the recording of a call. He anonymizes it first, that's the fixed first step, not an option, then distills it into a pattern register: the questions my clients ask again and again, each paired with my own best answer to it. From that register come a playbook, FAQ answers, raw material for content, and a library of objections. What he doesn't deliver: a new opinion. He distills what I have already said myself.

Join the communityRead first: Learning Claude

A worked example from the course

Numbers from the course lesson, not a figure of my own: a worked example of how much expertise sits unused inside conversations.

300 calls, 45 min. each

Worked example in the course

225 hours

Expertise distilled from it

Batches of 10 to 20 calls

Work step

Before any processing happens, a fixed anonymization pass runs: names become a role, companies become an industry, amounts become an order of magnitude, places become a region, and sensitive passages about health, family or private finances are removed, not replaced. The reason isn't excess caution, it's the law: secretly recording words spoken in private is a criminal offense under German law (§ 201 StGB), not a minor infraction. The course supplies three practical consent paths with templates, before a microphone ever runs.

The actual processing runs in batches of 10 to 20 calls, with a sample review after every batch, and time spent is measured on the first batch rather than estimated in advance. The 225-hour figure above is a course example for illustration, not a number from my own records: any existing recordings without consent stay out entirely regardless, more on that in the next section.

The tools

Gustav has his own 13-lesson course in my community that shows exactly this chain:

  • Whisper runs locally on the machine itself, only the anonymized extracts later go into cloud processing, the raw transcript never leaves the machine.
  • A fixed anonymization pass before any processing: role instead of name, industry instead of company, order of magnitude instead of amount, region instead of place.
  • Processing in batches of 10 to 20 calls with a sample review after every batch instead of one giant run.
  • A ledger that always shows which call belongs to which person, so a deletion request can actually be carried out.

What he can't do, where the line is

  • Without a clean legal basis, Gustav doesn't get to see a transcript at all. Consent first, then the recording, then the processing.
  • Anonymize first, process second, always. That's the fixed first work stage, not an option.
  • Existing recordings without consent stay out: either consent is obtained afterward, or the call stays private or gets deleted. There is no retroactive fix.
  • Your own answers, not AI wisdom: Gustav distills what I have said myself, and doesn't invent advice on top of it.
  • A deletion plan is mandatory, not optional: the ledger always shows which call belongs to which person.

The path: build it yourself or load the template

Standard: build it yourself

The entry point is my learning path: Learning Claude, then the process for hiring an employee, then the actual build. What's special about Gustav is three building blocks: anonymization as a fixed first stage instead of an afterthought, the pattern register instead of the raw transcript as the actual product, and batch review instead of trusting a single giant run.

Premium: ready-made template

Gustav is a ready-made package in the community: Standard ($9 a month) means building it yourself, Premium ($45 a month) means loading the tested template and adapting it, including a fictional demo transcript to practice on. As of September 19, 2026, prices in US dollars per Skool.

Frequently asked questions about Gustav

Can AI write a meeting protocol?

Yes, a protocol is the simplest level: who said what gets recorded and summarized. Gustav doesn't stop there: he distills the recurring questions across many calls and my own best answer to each one, instead of just retelling a single conversation.

Can conversations be transcribed with AI in a GDPR-compliant way?

Only with a clean legal basis in place beforehand: secretly recording words spoken in private is a criminal offense under German law (§ 201 StGB). Gustav's course supplies three practical consent paths with templates, then the recording runs locally via Whisper, and a fixed anonymization pass comes before any further processing.

How do you transcribe conversations automatically?

Through a local speech recognition tool like Whisper, which turns the recording into text without the raw data ever leaving the machine. Only after that does Gustav's actual work start: anonymize, then distill into a pattern register of recurring questions and answers.

Does Whisper really run locally, without the cloud?

Yes, in the course Whisper runs directly on the machine itself, the raw transcript never leaves it. Only once the fixed anonymization pass has removed or generalized names, companies, amounts and places do the anonymized extracts go into cloud processing for the pattern register.

What sets AI meeting notes apart from a real analysis?

Meeting notes summarize a single conversation. A real analysis like Gustav's pattern register compares many conversations against each other and shows which question keeps coming back and which answer worked best, as a basis for a playbook, FAQ and content, not as the record of one call.

Read on

Want to know what's hiding in your own calls?

Gustav's template is in the community, with the anonymization pass and consent templates ready to rebuild.

Join the community