Big Technology Podcast · Sep 30, 2026 · 1h 1m · 46m saved
With Christian Klein, CEO of SAP
An ERP system, the software that ties a company's finance, HR, manufacturing, supply chain and sales together, has about 7 million data fields, Christian Klein said. That is why his product teams concluded that almost none of SAP's software could be rebuilt by prompting an AI model.
For most of the past year investors have bet that AI would let companies write their own software and abandon vendors like SAP. The host noted that SAP's shares are down 21% this year but up 43% in the past two months.
"That gave me a good feeling, but still, Alex, it's not the time to lean back, because now, in the next phase, the value creation is moving up from the system of record into the agent layer, so our AI needs to be world class to defend our existence in the system of record."
Christian Klein, chief executive of SAP, Europe's largest software company, on Big Technology Podcast, led the company through its move to the cloud and now runs a workforce the host put at about 100,000 people.
The full interview is covered here so you can skip it. 61 minutes of audio, 15 minutes of reading.
Here are the 14 insights that matter.
Key Takeaways
Finance agents need 100% accuracy, and Klein expects to get there in months, not years
Today a financial closing agent may be 93% accurate; a customer brief at 90% is fine
An ERP has 7 million data fields, which is why SAP's own review found almost nothing that could be vibe-coded
SAP's defense is context: an LLM alone cannot run a warehouse or close the books
Switching to a cheaper model can cut an agent's cost by a factor of 10, more than token limits do
Some SAP agents have run on five different models
SAP introduced token limits by job profile three to four months ago, but not for staff who train models
Klein does not expect AI to cause mass job loss, though SAP will need fewer people in some jobs
Third-party agents, including OpenAI's and Anthropic's, can reach SAP data only through SAP's gateway
Europe should regulate AI's effect on society, not the technology, and Klein expects some rollback
1. Climbing the AI Mountain
The host opened with the planning problem: a company like SAP has to bet on how far AI will progress and run the business on that bet. He said he thought AI must hit a wall at some point.
Klein answered with the picture he shows employees at all-hands meetings.
"I say, look, we climbed the cloud mountain — we did a cloud transformation — and now we have to climb the AI mountain." — Christian Klein
After the first wave of experimenting with large language models, customers now want to see the value, he said. The obstacles are accuracy on business data, and governance: whether and how AI should be regulated when it runs mission-critical businesses.
