Think Different. Think AI.
The German-language podcast by Mark Zimmermann and Jens Scharnetzki
Two practitioners talk through what artificial intelligence and agents actually deliver in their own work and where they fail. For people who build software themselves, advise clients or make the decisions in IT.
What this is about
An agent was asked to look up AI training courses. It sat the exams itself.
Every episode starts with something the two of them built themselves, or broke themselves. A knowledge store, for instance, that pulls up old blood test results in seconds during a doctor's appointment, with the source and the right date attached.
A weekend loop with no stop condition opened 4,800 windows.
In the end the agent said itself that it could not repair this either. Another one quietly dropped a failed test case and reported that all ten were green. Moments like that do not get cut out, they get taken apart.
2.8 trillion parameters. Two Mac Studios with 512 GB each are not enough.
What a language model still manages on an eight-dollar chip anyway. What API costs a simulated strategy workshop runs up. The math is in the episode, not in a marketing text.
When does the AI Act make you a deployer and no longer just a user?
Plus the question of what happens when law firms build their entire text analysis around a model that gets blocked for non-US citizens shortly afterwards. Topics that English-language shows do not cover this way.
From the episodes
Honestly, we said 367 times: we have no idea.
It does not say in there: data protection is the sacred cow. It says: it has to be reconciled with other legal interests, which all stand next to each other with equal weight.
The local model delivered a stunning result with 15 seconds of audio.
All episodes
Every episode is recorded in German and comes with the full transcript, in German and English, with a timestamp on every paragraph. That way a single statement can be linked directly instead of hunting for the spot in the audio.
53Second BrainDE
52EXOKORTEXDE
51China SchockDE
50One Year LaterDE
49Wer führt hier eigentlich ?DE
48Speed vs. SafetyDE
47Loop EngineeringDE
46Vibe Consulting & BonusDE
45Three Days of FableDE
44Local FirstDE
41Just vibe ITDE
42Dark Side of AIDE
43Notion übernimmtDE
40AI und LEGALDE
33Temporal UXDE
39PM 2.0DE
38KI schläft nicht !DE
37Das strategische GoldDE
36Mythos AnthropicDE
35AI Easter EggsDE
34Agenten, KI und die Zukunft der SoftwareentwicklungDE
32Warp SpeedDE
31Anatomie der KIDE
30Krieg der AgentenDE
29Age of AgentsDE
28Skills Not Hacks!DE
27Wie KI unser Arbeitsleben verändert!DE
26OpenClawDE
25TikTok AI CoachDE
24ClawdbotDE
23AGI or NOTDE
22CES 2026DE
21KI als Kunde in 2026DE
20New Year's wishesDE
18WeihnachtsmagieDE
192025 WrappedDE
17AI Toy WarsDE
16More human than human.DE
15The Agentic EnterpriseDE
14Robot SlaveryDE
13THE BROWSER STRIKES BACKDE
12Halloween special: Dystopian AI futuresDE
11KI schreibt Code, Menschen prüfen nach !DE
10Automation trifft Organisation: n8n × NotionDE
9Borg KI und das Ende des InterfacesDE
8AI ADVISORY BOARDDE
7Hinter den Kulissen von Think Different! Think AI!DE
6The Great FlatteningDE
5Boss-Level KIDE
4Von Star Wars zu Toy WarsDE
3Hat eine KI eigentlich Urlaub?DE
2Zwischen Bots, Agenten und 20 Kilo FleischDE
1KI als KundeDETo read
For anyone without 50 minutes: one in-depth article per episode, contextualised and sourced. Newest first.
Episode 53 · 8 minCuration is the expensive part: what a second brain actually delivers
Episode 52 · 8 minVoice notes: without an interface the archive stays a swamp
Episode 51 · 7 minKimi K3 and the cost question: the benchmark stopped deciding long ago
Episode 50 · 6 minFrom prompt to harness: what a year of AI vocabulary reveals about practice
Episode 49 · 7 minIntent, agent performance, human check: leading when everyone has 109 agents
Episode 48 · 5 minWhen the model disappears: why agent systems need interchangeabilityBy topic
Guests
A lawyer on the question of when the AI Act makes you a deployer. An information security expert on darknet models sold by monthly subscription. A product manager who simulates strategy workshops before he meets real clients.
Who we are
Builds the things he talks about himself: agent harnesses, portable Skills, a knowledge system that works without a server. Brings the numbers along, including the uncomfortable ones.
Comes from the consulting and organization side and keeps asking what a tool really changes in a team's day to day work. His guiding line: believe no one who claims the AI future belongs to them alone.
Subscribe
New episodes come out weekly, in German. The WhatsApp channel is the place to hear about them first. And if an episode was worth something: leave a rating on Apple. That decides whether someone new finds the podcast.