AI · Nov 9, 2026
OpenAI’s Control Room puts a fail-closed kill switch inside the inference engineSAP and Infosys say enterprise agents need data plumbing and governance first. They also sell both
· Updated · Signal Desk newsroom
Analysis of reporting by MIT Technology Review. How we work

A sponsored MIT Technology Review Insights podcast with SAP and Infosys executives argues agents fail without connected data, clear outcomes and governance. The advice is sound; the framing comes from vendors with a stake in it.
Key takeaways
- This is a Business Lab episode produced in partnership with Infosys by MIT Technology Review’s custom content arm, not its editorial staff.
- SAP’s Thorsten Leiduck says AI is “absolutely dependent” on a connected, semantically rich data foundation.
- He argues 80% accuracy may suffice for everyday tasks but is unacceptable in supply chains or financial close.
- Infosys’s Satinder Pal Singh proposes judging agent use cases on impact, feasibility and scalability, and warns missing governance creates a “risk backlog.”
- Both foresee agents coordinating work across applications, with knowledge graphs and governed data as key assets.
Start with the disclosure
This piece is sponsored. The episode of MIT Technology Review’s Business Lab was produced in partnership with Infosys by the publication’s custom content arm, and features an SAP senior vice president and an Infosys associate vice president who works on SAP AI strategy. When two companies whose businesses include enterprise data platforms and implementation services say the key to AI is a better data foundation and stronger governance, readers should weigh that accordingly. That doesn’t make the advice wrong. Much of it matches what independent observers have said about why enterprise AI pilots stall.
The core argument
SAP’s Thorsten Leiduck frames the shift as moving from a pile of disconnected AI experiments to AI as an enterprise platform. His central claim is that AI is “absolutely dependent” on a connected, semantically rich data layer that reflects how the business actually works. His most useful line concerns accuracy thresholds: for some everyday applications, he says, 80% might be good enough, but in supply chains or financial close that is a terrible result. That distinction is exactly what many agent pitches skip. An agent that drafts emails can be wrong one time in five; an agent that books journal entries cannot.
Infosys’s Satinder Pal Singh calls agents “digital coworkers” and offers a three-part filter for use cases: impact on meaningful business metrics, feasibility given the company’s data and talent, and scalability. His line that a pilot that can’t scale is just an experiment is a fair critique of how many companies have spent the last two years. On governance, he argues that skipping policies, access controls, monitoring and oversight doesn’t speed adoption; it builds a risk backlog that eventually forces a disruptive stop-and-audit.
Where we agree, and where we don’t
The governance point is the strongest and the most timely. As agents gain the ability to act across systems, the cost of an unmonitored mistake rises sharply, a theme in another recent Technology Review piece on fencing AI in rather than predicting it. Companies that build access controls and audit trails before scaling will move faster later, not slower.
The forward-looking claims deserve more skepticism. Singh describes a future where AI becomes a “decision-making partner,” with a CFO’s agent surfacing the three issues most likely to affect the quarter. He also says use cases that look constrained at program start are often resolved before go-live, and advises designing for where AI will be in 18 months. That is optimistic, and conveniently argues for large platform commitments now. Other enterprise evidence has been more mixed; earlier reporting found agent impact on many businesses still small. The data-foundation thesis also has a self-serving edge: it implies the path to AI runs through consolidating data in the right vendor’s platform.
Who should listen
- CIOs and CFOs
- A useful framework on accuracy thresholds and use-case selection, with vendor incentives to discount.
- Risk and compliance teams
- Strong support for building governance before scaling agents, not after.
- SAP customers
- A preview of how SAP and Infosys will pitch agent deployments and knowledge graphs.
- Agent startups
- Evidence that incumbents will argue data platform depth beats standalone agents.
What to watch
- Customer case studies with measured outcomes from SAP and Infosys agent deployments.
- How enterprises set accuracy thresholds for agents in finance and supply chain processes.
- Governance tooling, including access controls and audit trails, built into agent platforms.
- Whether knowledge graphs become a standard prerequisite in enterprise agent projects.
FAQ
Is this independent journalism?
No. It is a Business Lab episode produced in partnership with Infosys by MIT Technology Review’s custom content arm, not its editorial staff.
What is the main argument?
That enterprise AI agents depend on a connected data foundation, clear business outcomes and governance built for scale.
Why does accuracy matter so much?
SAP’s Leiduck says 80% accuracy may be fine for everyday tasks but is unacceptable in critical processes like financial close.
How should companies pick agent use cases?
Infosys’s Singh suggests weighing impact, feasibility and scalability.
More on AI
AI · Nov 11, 2026
OpenAI’s $2 million for children’s AI safety research matters for one promise: publishing bad newsAI · Oct 28, 2026
RTX Spark laptops get open speech models and a dial for background agentsAI · Nov 17, 2026
OpenAI says its models now shave area off production chips, including AMD’s next flagshipsAI · Nov 5, 2026
NVIDIA’s four open Cosmos 3 Nano models let robot and driving teams simulate worlds that react