DE

Workshop Series Finished

Nostalgie und KI. Eight episodes, one ear.

Can a machine tell history in a way that makes you believe it?

From the workshop

episodes, 2:43 hours in total, plus a trailer
8
versions until sign-off; 159 mixes in the production folders, none deleted
32
notes on the timeline of the review page, 92 implemented
93
factual errors among 560 checked claims in four scripts, found before I listened
83
my time: 18:49 h measured on my turn, plus 50% for listening
28:13 h
AI cost, not counting running subscriptions or the ElevenLabs overage charge
≈ €88.66

What it cost

Claude Code as assistant share of the Max plan; about 5.9 billion tokens, ≈ €2,199.32 at API list price ≈ €43.61
Claude Code via extra usage part of the dashboard work, after the plan allowance was used up, billed at API price; derived from the account balance, range €8.58 to €40.57 ≈ €18.63
Narrator, scenes, trailer, pronunciation tests (ElevenLabs) upper bound: 378,827 characters in the production log at the Creator rate (22 US dollars for 300,000 credits); the credit counter showed clearly less. Not included: the overage charge beyond 300,000 credits and two attempts with Audio Isolation ≈ €23.89
Ambience and sound effects (ElevenLabs Sound Effects) 31 ambience tracks and sounds, 465 seconds in total; the log shows no credits for them not quantified
Music (Suno) 24 clips filed, 14 of them in the episodes; a first pass discarded; opener and closing credits from the music bed set in plan
Cover images (Higgsfield) 15 images with Nano Banana Pro in 4K, 4 credits each; 10 in use ≈ €2.53
Hosting and feed existing web hosting, a dedicated subdomain for the audio files €0
Tools on my own machine Whisper, wav2vec2, Silero VAD, ffmpeg, Python €0
Total≈ €88.66

Effort per round

RoundClaude at API priceElevenLabsMy time (+50%)Machine
Concept, episode 1 in six versions, review page, series music≈ €37.11≈ €2.373:56 h2:08 h
Episodes 2 to 4≈ €63.55≈ €4.403:52 h2:41 h
Episodes 5 to 8 with fact-check≈ €242.95≈ €5.691:36 h2:40 h
New mixing software and blind listening test≈ €326.58–1:27 h9:06 h
All eight episodes remixed after the listening test≈ €472.61≈ €0.312:32 h7:33 h
Note round with 62 notes, breath checker≈ €487.74≈ €9.421:40 h11:02 h
Individual sign-offs up to the full season≈ €46.84≈ €1.535:47 h2:03 h
Release: website, feed, trailer, cover, platforms≈ €473.70≈ €0.185:54 h9:56 h
Download counting and dashboard≈ €48.23–1:30 h1:27 h
Total≈ €2,199.32≈ €23.8928:13 h48:36 h

Claude figures at API list price for comparison; almost everything actually ran on the Max plan (a share of about 2%), only part of the dashboard work via extra usage. ElevenLabs by characters in the log at the Creator rate, as an upper bound. My time: measured time on my turn, each break capped at 30 minutes, plus 50% for listening; parallel sessions counted once. Not included: this report.

The trailer: eight stories from the past and today’s AI 1:24
The trailer as it runs on the platforms, in German. Narrator: my voice as a clone. Scenes with library voices from ElevenLabs. Music: MF 01 and NK 02, made with Suno. AI GENERATED · KI-generiert.
Episode 1 · The message, 1991 0:41
The cold open: Gisela Behrens leaves her son a message on the answering machine and in the end doesn’t know how to switch the thing off. Library voices from ElevenLabs, telephone and cassette sound as filters. NK 02, made with Suno, fades out at the start. The Behrens family is fictional. AI GENERATED · KI-generiert.

The idea

I wanted a feature series with drama scenes, thirty minutes per episode, with me as narrator. The title was already set: “Nostalgie und KI” (Nostalgia and AI).

Each episode takes an object from the past and follows it to the present. The answering machine leads to the cloned voice, the Walkman to the AI companion. The assistant suggested the subtitle, “Jede Zukunft hat eine Vergangenheit” (Every future has a past), and I kept it.

After a few clarifications I wrote “okay - go for it. full shot. no risk no fun.” My first direction on the first version: the scenes drag, one answering machine is enough.

The format

The narrator is my clone, the ichbinsprecher.de brand voice, spoken through ElevenLabs Eleven v4. My first note on the voice was about its pressure. Out of eight test variants, one stage direction before every paragraph survived, [relaxed, warm, understated], which lowers the tone.

Most scenes are played by the Behrens family, and it is fictional. Gisela, her husband Werner, her son Thomas and his daughter Lina carry the series from 1972 to 2026. A “Fräulein vom Amt” (the switchboard operator of old German telephone exchanges) was also planned as a recurring framing character. In the second round of notes I wrote: “Fräulein raus… voll doof.” (Drop the operator… totally lame.) Since then the narrator gives the years.

Several episodes play the same situation twice, once then and once now.

Episode 2 · Parents’ evening, 1978 0:33
The pocket calculator arrives in class tests, and Gisela and Werner Behrens ask questions. The dialogue was generated in a single pass with ElevenLabs. NK 02, made with Suno, fades out at the start. AI GENERATED · KI-generiert.
Episode 2 · Parents’ evening, 2026 0:35
The mirror scene, almost fifty years later. Same sequence, now the subject is AI in homework. AI GENERATED · KI-generiert.

After the title, every episode mentions in passing “Produziert mit KI.” (Produced with AI.), and the closing credits spell it out.

Eight episodes, 32 versions

EpisodeVersionsLengthWhat triggered the rounds
1 · Answering machines and AI voices1021:1031 notes in the first review round, dedicated music, one sentence at the end
2 · Pocket calculators and ChatGPT622:30v1 ran eleven minutes; second research pass, teachers, watermarks
3 · Brockhaus (the German encyclopedia), Wikipedia and AI answers421:12context right after the scene, “ein Meter siebzig” (one metre seventy)
4 · Video store, Netflix and AI films423:04two content requests: one more scene, TV on demand
5 · Slide-show evenings and AI photos218:13three notes
6 · Quelle, Neckermann (German mail-order houses) and AI agents217:48five notes, serious music for the Nazi-era history
7 · Walkman, AI music and AI companions220:3218 notes: Walkman, content warning, AI music
8 · The millennium and AI forecasts218:43two notes

Episode 1 needed ten versions, episodes 5 to 8 two each. In between lies a rulebook. After the first review I wrote: “Mir ist wichtig, dass ich mich dem perfekten Podcast nähere” (What matters to me is getting closer to the perfect podcast). The assistant turned my 31 notes on episode 1 into rules and a checklist that every new episode is checked against before my review.

My goal was thirty minutes; I signed off eighteen to twenty-three. Length comes from content, never from padding, says the rulebook.

The truth in the text

Language models tell stories fluently, even where they are wrong. So every script carries a source table, 340 entries in total.

Episodes 5 to 8 went through a fact-check before I heard a single line. It ran as checking tasks for the assistant, so-called agents. For each episode, two checked the facts against sources on the web and one checked the rulebook. Every factual finding was cross-checked by two more, which adds up to 196 agents. Among 560 checked factual claims they found 83 errors: 29 wrong, 41 overstated, 13 unsupported. The script for episode 6 said Neckermann’s first catalogue was from 1950. His textile mail order was already running in 1937/38, through Karl Joel’s “Aryanized” company.

Then the real recordings. Episode 1 plays recordings from 1860, 1901 and 1927. The idea of cleaning them up with AI was mine. The assistant built in a cleaned-up version of the 1860 recording. I listened and wrote: “Na - ganz offenbar hat die KI was neues erfunden hier… Wir streichen das Experiment” (Well, the AI has obviously invented something new here… We’re dropping the experiment). The rulebook has its own line for it. No experiments in which the AI invents something.

The review

For the review I asked for a dedicated page. A player on whose timeline I place notes. The assistant replies at the same spot. There were 93 notes, 92 implemented, one discarded.

In the big round I placed 62 notes across all eight episodes. 23 were about the music, mostly its fades. At some point I only wrote one word, “Blendenthematik” (the fade issue). 17 concerned pronunciation, eight breaths and edits, eight the pause before the closing credits, five the content. One was praise.

Eleven notes were about a single word. My clone said “Wok-Man”, a German reading of the word. At the third instance I wrote “OMG WALKMAN NICHT WOK-MAN” (OMG WALKMAN NOT WOK-MAN). Of the solutions the assistant offered, I chose saying it less often and checking strictly. The word now comes up nine times instead of sixteen. Every remaining instance has to sound English in the phoneme check, in at least seven out of eight measurement windows.

v1 · “Walkman”, first version 0:30
“Walkman” comes up at 0:06 and 0:25. My ear flagged both as “Wok-Man”. Music: NK 03, made with Suno. AI GENERATED · KI-generiert.
v2 · After eleven notes 0:28
One “Walkman” fewer; the remaining one at 0:23 is English and confirmed by the phoneme check. Music: NK 03, made with Suno. AI GENERATED · KI-generiert.

Breaths were another recurring theme. On an in-breath at the end of a take I noted: “Das wirkt wie ein Schnittfehler.” (That sounds like an editing error.) When I asked whether breaths can be detected, the assistant built a breath checker. It found all seven spots I had marked, and two I had missed. Across 682 take boundaries it raised no false alarm.

v8 · A breath leading nowhere 0:23
At 0:17, after “angelegt hatte.”: the voice breathes in, then the paragraph pause follows. Music: NK 03, made with Suno. AI GENERATED · KI-generiert.
v9 · Breath removed 0:23
The take is silent from the end of the word; the pause keeps its length. Music: NK 03, made with Suno. AI GENERATED · KI-generiert.

The last version of episode 1 was made for a single sentence. In the outlook, the assistant had cut a wrong year out of the take, and the sentence lost its ending. I heard it at the very end, nothing else. Six new takes were made for v10, chosen by measurements.

v9 · Outlook with an edit 0:24
The wrong year 1976 is cut out of the take; episode 2 is set in 1978. At 0:11 “Elternabend” (parents’ evening) ends without closing the sentence, followed by an inserted pause. Music: MF 08, made with Suno. AI GENERATED · KI-generiert.
v10 · The same sentence, spoken again 0:24
Instead of the edit, a new take, chosen from six candidates. The voice drops after “Elternabend” and closes the sentence. The signed-off version. Music: MF 08, made with Suno. AI GENERATED · KI-generiert.

The check before I listen

One of my shortest briefs read “setz das alles um bitte vorher will ich nicht hören” (please implement all of it, I don’t want to listen before that). The production followed it. Before a version reaches the review page, a chain runs.

  • Speech recognition compares every take with the script.
  • A phoneme check listens for names and problem words such as Margenfeld and Walkman.
  • An intensity check has outlier takes regenerated.
  • The breath checker looks for in-breaths leading nowhere.
  • A music checker measures the fades and holds −16 LUFS. All eight final versions pass it without errors.

The mixing software and checking programs in the shared library come to a little over 9,000 lines of code.

The music

All the music comes from Suno. Opener and closing credits come from my music bed set, plus seven series themes. Piano, vocals, brass and anything heroic are excluded.

I liked the sound; the fades felt clumsy. As a benchmark I named a documentary series on OpenAI by Deutschlandfunk (German public radio), and the assistant measured it internally. There the music sits in blocks of a little over a minute. In my versions it rose in every pause between sentences, 9 to 15 times a minute.

This led to new mixing software in 15 versions. I decided the direction blind, in six listening comparisons. Four times I chose differently from the proposal: one music bed per chapter instead of music with dry stretches, a more present bed, the first paragraph of “Was bleibt” (What remains) without music, and the old ending. The bed now sits 17 LU below the voice. The plan was 23; the reference suggested 20. In the final versions the music now rises only 0.7 to 2.3 times a minute.

Under the original recordings in the teaser, the music no longer fades out; it only dips.

v8 · Teaser with the 1901 recording 0:40
Before the emperor, the music fades out; the original recording only follows after a 2.8-second pause. Original recording: Emperor Franz Joseph I, Vienna 1901, Wikimedia Commons, public domain. Music: NK 02, made with Suno. AI GENERATED · KI-generiert, except the original recording.
v10 · The same teaser, signed off 0:37
The original recording follows after half a second (0:17). The music keeps running underneath, about 5 dB quieter, and comes back softly afterwards. Music: NK 02, made with Suno. AI GENERATED · KI-generiert, except the original recording.

Where the subject is death, grief, suicide or Nazi crimes, the music falls silent. For the Nazi-era history of the mail-order business I relaxed the rule: “wir dürfen schon Betten hier verwenden… halt kein lustiges.” (we can use beds here… just not a cheerful one.) Following my note, episode 7 carries a suicide notice in the opener, with the number of the Telefonseelsorge (the German crisis helpline).

Release and measurement

The recommendation was to launch with the trailer and three episodes. After my “GO GO GO” all eight went online at once. The AI notice is also in the feed in machine-readable form. On the episode pages, the scripts’ source entries are resolved into 570 individually linked references. Downloads are counted without cookies. The first count showed nine downloads, seven of them my own test requests.

What still sounds like amateur dramatics

About the scenes I once noted “etwas Laientheatermäßig” (a bit amateur-dramatics). The library voices can reach dubbing quality only to a limited degree, the assistant replied. I never reached the thirty minutes either. What holds the series together is something else. Its facts are cross-checked, its original recordings untouched.

A look at the case

Can a machine tell history in a way that makes you believe it? After this season I would say: if it distrusts itself. Agents found the factual errors, and a cross-check confirmed every single one. What no check knew about, the ear found. A Wok-Man, one breath too many, a fade too short. Each of these notes became a rule, some became a checking program.

At API price, concept, scripts and fact-check cost about €344; sound and polish afterwards about €1,334. The truth was cheaper than the sound. What I actually paid for the assistant was about €62, because it ran on the plan; with voices and images, about €89.

What this means for an organization

Strategy. The first draft has become cheap. About four fifths of the assistant cost up to sign-off came after it. A budget measured against the first draft would have been almost five times too small here.

Role. It takes a cross-check, and a machine can do it. The rules it checks against are set by a human. Here there were two: every factual finding is cross-checked independently, and nothing is invented, not even to improve something. The second came from an idea of my own that I dropped.

Process. Note, rule, checking program. The notes on episode 1 became a rulebook, the breath notes a checking program. What the ear has found once, the machine looks for next time. All 159 mixes in the production folders are still there, including the discarded ones.

What stays human. The limits. Operator out, AI restoration out, content warning in, serious music for the Nazi-era history, nothing goes out without asking. After the first review I wanted to get closer to the perfect podcast. Three versions later I wrote “folge 1 nahezu perfekt” (episode 1 nearly perfect). The “nearly” is something a human hears.

Division of labor

Who did what?

Human

  • The idea, the title and the format: a feature with drama scenes, me as narrator
  • The review page as a brief, and 93 notes on its timeline
  • The verdict in the blind listening test: one bed per chapter, more present than planned
  • The limits: switchboard operator cut, AI restoration cut, content warning in episode 7
  • Every sign-off, the launch date and every “go” for the release

Machine

  • Research, scripts with 340 source entries and 46 scenes, following a rulebook built from my notes
  • The fact-check with 196 agents: 83 errors found, cross-checked and corrected
  • Narrator, scenes and ambience via ElevenLabs, every take measured and selected
  • The new mixing software in 15 versions, breath checker and phoneme check
  • Feed, episode pages, transcripts, upload and the platform listings

Tools

  • Claude Code (Opus)
  • Claude in Chrome
  • ElevenLabs
  • Suno
  • Higgsfield
  • Nano Banana Pro
  • Whisper
  • wav2vec2
  • Silero VAD
  • ffmpeg
  • Python

Notes

All voices, scenes, ambience, music and images are AI-generated, including my own voice. Only the historical original recordings in episode 1 are real.
Idea, title, direction and sign-off: Stefan Margenfeld. Narrator: the ichbinsprecher.de brand voice as a clone (“Ichbinsprecher.de - Imagefilm-Tonality”, ElevenLabs Eleven v4). Scene voices from the ElevenLabs library, among them Dana (Gisela), Marc (Werner), Ben and Peter Hartlapp (Thomas), Mila Winter (Lina), Susi (AI voices in the scenes). The Behrens family is fictional. Ambience: ElevenLabs. Music: MF 01 and MF 08 from the Margenfeld music beds, series themes NK 01 to NK 07, made with Suno. Original recordings in episode 1: Scott de Martinville 1860 (transfer: First Sounds, firstsounds.org, CC BY; shortened), Thomas Edison 1927 (Thomas Edison National Historical Park, National Park Service), Emperor Franz Joseph I 1901 (Wikimedia Commons, public domain). Cover image: Nano Banana Pro via Higgsfield. Claude Code as a share of the Max plan (200 US dollars a month), calibrated against the measured weekly allowance; comparison at the API list price of Claude Opus 5.5. ElevenLabs at the Creator rate, Higgsfield at list price (1,000 credits = 49 US dollars). Rate €0.86 per US dollar.

How I measure →

Quick questions

Frequently asked questions

What is the podcast “Nostalgie und KI”?

“Nostalgie und KI” (Nostalgia and AI), subtitled “Every future has a past,” is a German-language podcast series by Stefan Margenfeld with eight episodes, 2:43 hours in total. Each episode follows an object from the past to today’s AI: the answering machine, the pocket calculator, the Brockhaus encyclopedia, the video store, the slide show, the mail-order catalog, the Walkman and the turn of the millennium. The narrator is Margenfeld’s voice clone, and the drama scenes feature the fictional Behrens family. The series is on margenfeld.com/podcast/nostalgie-und-ki/ and the major podcast platforms.

How were the facts in “Nostalgie und KI” checked?

Every script keeps a source table, 340 entries in total. For episodes 5 to 8, AI agents checked the facts against sources on the web, and two more agents cross-checked every finding; 196 agents in all. Among 560 factual claims they found 83 errors: 29 wrong, 41 overstated, 13 unsupported. All were corrected before Stefan Margenfeld heard a single line.

What did the podcast series cost, and how much work went into it?

AI costs came to about €88.66, excluding running subscriptions and an additional ElevenLabs charge. At API list price, the assistant would have cost about €2,199. Stefan Margenfeld’s own effort was 28:13 hours. Thirty-two versions of the eight episodes were made before sign-off.

How do human and AI work together during sign-off?

Stefan Margenfeld listens on a dedicated review page and places notes on the timeline; the assistant replies at the same spot. There were 93 notes, and 92 were implemented. Whatever the ear finds becomes a rule and sometimes a checking program: the notes on breaths led to a breath checker, and “Wok-Man” instead of “Walkman” led to a pronunciation check.

What can organizations learn from “Nostalgie und KI”?

The first draft has become cheap. About four fifths of the assistant costs came after the first draft; a budget based on it would have been almost five times too small. A machine may do the cross-checking, but a human sets its rules. Stefan Margenfeld’s are: every factual finding is checked independently, and nothing is invented, not even to improve something.

What role does Stefan Margenfeld play in his AI productions?

Stefan Margenfeld sets the idea, format, limits and standard, and signs off every version. On “Nostalgie und KI,” for example, he dropped an AI restoration of historical recordings because the AI invented something in the process, and he added a trigger warning to episode 7. Research, draft scripts, voices, mixing and checks come from the machines. He makes this division of labor visible in every workshop report.

Keep thinking · From the journal

What future do we want to build together?

Let’s talk. Next workshop notes: THE AGENCY