A Beginner's Guide to AI

Agentic finance is changing how companies are analyzed. AI can now examine hundreds of pages of corporate disclosures, compare years of financial data, and identify patterns that human readers may miss.

Michael Diegelmann joins Dietmar Fischer to discuss Warren Wise, a multi-agent financial-analysis system based on principles associated with Benjamin Graham, Warren Buffett, and Charlie Munger.

Michael explains how CFOs and investor-relations teams can test annual reports, earnings materials, and equity stories before publication. The system can identify potential weaknesses, generate difficult investor questions, compare companies through a consistent methodology, and show how banks or institutional investors may interpret the information.

The conversation also explores the limits of financial AI. A professional system must distinguish between a missing value, a dash, a negative number, and an unusual human event. Without reliable data, guardrails, and context, a convincing AI answer can still be wrong.

🔍 In this episode:
What agentic finance means for companies
How AI reads annual reports and financial footnotes
Why recurring EBITDA adjustments may become warning signals
How AI can examine links between performance and management compensation
Why CFOs need to anticipate machine-generated investor judgments
How financial AI differs from a general-purpose chatbot
Why human oversight remains essential
Which finance professionals risk being left behind

Michael argues that AI will make corporate finance faster and more transparent. It may also give executives an independent way to challenge internal assumptions before the market does.

Get the Newsletter!
📧💌📧
Tune in to get my thoughts and all episodes. Don't forget to subscribe to our newsletter:
https://beginnersguideto.ai
📧💌📧

About Dietmar Fischer
Dietmar Fischer is a podcaster and AI marketer from Berlin.
If you want help with AI strategy or digital marketing, visit:
https://argoberlin.com

Quotes from the Episode
“The thing is, you can't hide stuff anymore.”“There's a new sheriff in town, and this is called artificial intelligence.”“Everything may sound or look bad, and still isn't.”

Chapters
00:00 Meet Michael Diegelmann
01:59 Building a Synthetic Warren Buffett
05:09 Testing an Equity Story Before Publication
08:07 The Patterns Hidden in Corporate Reports
15:41 The Coming Tsunami of Transparency
18:08 Why Financial AI Needs Better Data and Guardrails
28:12 Agentic Finance, Jobs, and the Future of Corporate Leadership

Where to Find Michael Diegelmann
🌐 Warren Wise: warren-wise.com
💼 LinkedIn: Michael Diegelmann
🏢 cometis AG: cometis.de
Hosted on Acast. See acast.com/privacy for more information.

Sam Altman Just Admitted AI’s Favourite Metric Is Fake.
Sam Altman calls tokens a questionable unit, Amazon warns against tokenmaxxing, and enterprises search for a better way to measure AI ROI.

🪙📊 AI providers count tokens, price tokens and report tokens. Businesses then use those figures to track adoption, manage budgets and sometimes evaluate productivity. But token consumption measures computational activity, not whether the AI solved a useful problem.

In this episode of A Beginner’s Guide to AI, we examine the gap between AI token pricing and AI business value. Sam Altman has called tokens “a kind of fake unit,” while also explaining why customers prefer a visible and predictable meter. Paying for results sounds more logical, but outcome-based AI pricing requires clear goals, reliable benchmarks and a way to prove what the AI actually caused.

We also look at the rise and fall of tokenmaxxing. Amazon warned that token usage and AI adoption metrics become deeply flawed when they are separated from real results. The company did not reject usage data. It placed that data in context, next to useful software, reliable releases, lower costs and customer impact.

💡 In this episode, you will learn:
Why AI tokens measure activity rather than value
Why enterprises still prefer predictable token pricing
How tokenmaxxing can distort behaviour and increase costs
Why outcome-based AI pricing is difficult to verify
How to measure AI ROI beyond tokens
Why cost per successful task can be more useful than cost per token

📧💌📧
Tune in to get my thoughts and all episodes, don’t forget to subscribe to our Newsletter: beginnersguideto.ai
📧💌📧

Quotes from the Episode
“Tokens belong on the cost side of the equation. Outcomes belong on the value side.”“A perfectly efficient model can complete the wrong task with impressive economy.”“The real measure of AI success is not how much intelligence passed through the meter. It is whether something useful happened after the meter started running.”
About Dietmar Fischer
Dietmar is a podcaster and digital marketer from Argo.berlin. If you want to get your AI or digital marketing moving, contact him at argoberlin.com.

Sources for the case study and central quotations: Amazon engineering guidance on AI outcomes, CNBC reporting summarised by Quartz, Cheeky Pint transcript with Sam Altman and John Collison, and OpenAI’s AI value scorecard.

Chapters
00:00 The Token Trap
02:20 How Token Pricing Became the Default
06:55 Why Outcomes Are Hard to Price
10:40 The Cake That Ate the Budget
12:05 Amazon’s Tokenmaxxing Lesson
14:35 Measuring What AI Actually Solves
Hosted on Acast. See acast.com/privacy for more information.

The hidden cost of AI solutionism and what happens when optimization replaces human judgment.

🤖 AI can analyze data, rank people, predict risks and recommend decisions. But what happens when it solves the wrong problem?
This episode of A Beginner’s Guide to AI examines AI solutionism, the belief that complex human problems can be fixed mainly through better technology. The idea is attractive because technical problems can be measured and optimized. Social problems involve values, power, competing interests and uncomfortable choices.

AI does not remove those choices. It can hide them inside targets, datasets and risk scores.
A simple cake-sharing example demonstrates how an algorithm can produce a mathematically precise answer while missing the meaning of fairness. We then examine SyRI, an algorithmic system used by the Dutch government to identify possible welfare and tax fraud. The case shows what can happen when efficiency, surveillance and social policy collide.

A big thanks to the people of Airia:
Visit Airia to see how your organization can adopt AI securely, govern it effectively, and move beyond isolated chatbot experiments.

In this episode:
🔹 What AI solutionism means
🔹 Why algorithms can optimize the wrong objective
🔹 How automated decisions shift power and responsibility
🔹 Why accuracy is not the same as fairness
🔹 What SyRI teaches us about AI accountability
🔹 How to identify solutionism in your own AI projects
🔹 Why human oversight must begin before deployment

📧💌📧
Tune in to get my thoughts and all episodes, don't forget to ⁠⁠⁠⁠subscribe to our Newsletter at beginnersguideto.ai⁠⁠⁠⁠
📧💌📧

Quotes from the Episode
💬 “An automated decision is still a human decision with extra steps.”
💬 “The problem is not that the machine calculated badly. The problem is that we asked it to calculate the wrong thing.”
💬 “The most dangerous AI system may not be one that fails. It may be one that efficiently solves the wrong problem.”

Chapters
00:00 When Every Problem Looks Like an Algorithm
02:08 What AI Solutionism Means
07:42 The Perfect Cake Problem
09:35 SyRI: When Welfare Becomes a Risk Score
13:18 The Five-Question Solutionism Check
15:20 A Tool Is Not a Verdict

About Dietmar Fischer
Dietmar is a podcaster and digital marketer from Argo.berlin. If you want to know how to get your AI or your digital marketing going, just contact him at argoberlin.com.

Sources for the case study:
District Court of The Hague: SyRI legislation in breach of the European Convention on Human Rights and Library of Congress: Netherlands Court Prohibits Government’s Use of AI Software to Detect Welfare Fraud.

Thanks to our sponsor, Airia: Airia brings discovery, security, governance, and optimization together in one continuous platform.
Hosted on Acast. See acast.com/privacy for more information.

How divided government, state regulation, and local opposition to data centers could reshape America’s AI industry.
The US midterm elections delivered surprising results. But beyond the immediate political consequences, what do they mean for AI companies, data centers, and the future of American artificial intelligence?

In this solo episode of Beginner’s Guide to AI, I examine why sweeping national AI regulation may remain unlikely while political pressure moves to states, governors, utility regulators, and local communities.

Texas offers an important case. New restrictions on data-center development could respond to concerns about electricity, water, pollution, and property rights. This does not necessarily mean the end of new data centers. It could mean longer approval processes, higher costs, environmental requirements, and more serious negotiations with affected communities.

🏗️ Key questions in this episode:
Will divided government preserve the federal status quo?
Could state regulation slow the rollout of AI?
Should AI firms compensate communities for infrastructure costs?
Can local rights and national AI ambitions coexist?
Could stronger regulation give China a competitive advantage?
Did Silicon Valley lose the first real AI election?

My argument is that AI firms will need more than capital and technical capability. They will also need community support and a political license to build.

📧💌📧
Tune in to get my thoughts and all episodes. Don’t forget to subscribe to our newsletter: beginnersguideto.ai
📧💌📧

Quotes from the Episode
“The Congress can’t make a bill that the president won’t veto. So the status quo stays.”“The AI firms, they want their data centers. And if they want them, they have to go into community outreach.”“One party has the land and the energy and the water, and the other one has to just pay for it. And then it’s a market transaction.”“My bet is there will be stronger regulation, and one could say Silicon Valley lost the election. The AI election.”
About Dietmar Fischer
Dietmar Fischer is the creator and host of Beginner’s Guide to AI and a digital marketer at Argo.berlin. If you want to get your AI project or digital marketing moving, contact him at argoberlin.com.

Chapters
00:00 What the Midterms Mean for AI
01:19 Why National AI Regulation May Stall
02:33 Texas and the Data Center Moratorium
04:23 Why AI Firms Need Community Support
05:45 Regulation and the Race With China
06:37 Did Silicon Valley Lose the AI Election?
Hosted on Acast. See acast.com/privacy for more information.

Moravec’s paradox: 🤖 vs. 🧔🏻

AI can defeat grandmasters, generate code, and analyse complex documents in seconds. Yet machines still struggle with abilities that humans use without conscious effort, such as opening a door, handling a soft object, understanding everyday context, or tying a shoelace.

This is Moravec’s paradox: tasks that humans consider intellectually difficult can be relatively easy for computers, while perception, movement, and common sense remain some of the hardest problems in artificial intelligence.

In this episode of A Beginner’s Guide to AI, we examine the difference between producing an intelligent answer and acting intelligently in the physical world. We look at the AI grounding problem, embodied intelligence, physical AI, robot perception, motor control, and the hidden limits of automation.

A cake experiment demonstrates what happens when a robot follows its instructions perfectly but fails to adjust to a softer or slipperier object. We also examine the DARPA Robotics Challenge, where advanced robots attempted to drive, open doors, use tools, cross obstacles, and climb stairs in a simulated disaster environment.

The episode asks an important business question: Are companies testing AI under realistic conditions, or are they being persuaded by polished demonstrations? Strong performance in one task does not guarantee common sense, adaptability, or reliability when the situation changes.

Key Topics 🧠
Moravec’s paradox in AI
Why AI struggles with simple tasks
Human intelligence versus artificial intelligence
AI common-sense reasoning
Embodied AI and physical AI
Robot perception and motor skills
The grounding problem
Limits of AI automation
Lessons from the DARPA Robotics Challenge
AI reliability in unpredictable environments

📧💌📧
Tune in to get my thoughts and all episodes, don't forget to subscribe to our Newsletter: beginnersguideto.ai
📧💌📧

Quotes from the Episode 💬
“They receive a map made from words, while humans also walk through the territory.”“Intelligence is not a single dial that moves neatly from foolish to brilliant.”“The machine may know every written instruction for tying a shoelace. The toddler may still be the only one wearing properly fastened shoes.”

Chapters ⏱️
00:00 Why Toddlers Still Beat AI
03:05 Moravec’s Paradox and the Reversal of Difficulty
07:32 Common Sense, Grounding, and Human Experience
13:18 The Cake That Defeated the Robot
15:10 The DARPA Robotics Challenge
20:34 Finding the Intelligence We Usually Ignore

About Dietmar Fischer
Dietmar is a podcaster and digital marketer from Argo.berlin. If you want to know how to get your AI or your digital marketing going, contact him at argoberlin.com.
Hosted on Acast. See acast.com/privacy for more information.