Uploaded May 2026 | Updated September 2026, 2 weeks ago
AI can do the paperwork of decisions.
It still cannot own the consequences when supply chains fail.
I’m joined by Simon Bezrukov, Chief AI Officer at Bristlecone, to cut through the current noise around agentic AI, LLMs, digital twins, forecasting, simulation, and automation in supply chain operations. Simon’s work sits where AI stops being a slide and starts affecting service levels, inventory, planners, suppliers, cost, carbon, and risk. Handy, because the world has enough AI theatre already. It does not need better lighting.
This matters now because supply chains are being hit from every side: geopolitical disruption, cost volatility, sustainability pressure, labour constraints, ESG expectations, brittle data, and the ongoing fantasy that more automation automatically means more resilience. Simon makes a crucial distinction: automation is not autonomy. AI agents can fetch missing data, open tickets, draft supplier communications, or propose replans, but they struggle when decisions require unresolved trade-offs between service, cost, cash, carbon, and customer risk.
What changed my thinking was the clarity of the failure modes. LLMs are brilliant explainers, but they are not supply chain decision engines. More data helps until the world changes, and under real uncertainty machine learning can become confidently wrong. Simon also makes the case for smaller, well-governed models over perfect digital twins nobody trusts. Sometimes the best predictor is embarrassingly simple, like the length of a service request. Kismet: apparently, after all the algorithms have finished tap-dancing, counting words can still beat the clever machine. Humanity is safe for another Tuesday.
This is for senior supply chain, procurement, logistics, operations, manufacturing, sustainability, and risk leaders who need practical AI decisions, not expensive mythology.
If you’re applying AI to real supply chain decisions, I’d like to hear where you draw the line between automation and accountability.
🔗 Podcast website: resilientsupplychainpodcast.com
🎧 Subscribe or follow for more conversations on supply chain resilience, operational risk, sustainability, and decision-making.
⏱️ Chapters / Timestamps
00:00 – AI Can’t Own Supply Chain Consequences
01:15 – Turning AI Hype Into Supply Chain Metrics
02:49 – Why AI Fails Without Operational Objectives
04:08 – Agentic AI, Autonomy, and Fast Mistakes
07:52 – Where AI Breaks: Cost, Carbon, and Service Trade-Offs
11:32 – Why LLMs Are Not Decision Engines
19:08 – When More Data Makes Forecasting Worse
21:02 – Simulation When Disruption Breaks Forecasts
24:13 – Supply Chain Resilience Beyond Forecast Precision
25:39 – Why AI Amplifies Human Expertise and Bad Assumptions
28:23 – The Risk of Overbuilt Digital Twins
37:59 – AI Governance Before Live Operations Break
AI can do the paperwork of decisions.
It still cannot own the consequences when supply chains fail.
I’m joined by Simon Bezrukov, Chief AI Officer at Bristlecone, to cut through the current noise around agentic AI, LLMs, digital twins, forecasting, simulation, and automation in supply chain operations. Simon’s work sits where AI stops being a slide and starts affecting service levels, inventory, planners, suppliers, cost, carbon, and risk. Handy, because the world has enough AI theatre already. It does not need better lighting.
This matters now because supply chains are being hit from every side: geopolitical disruption, cost volatility, sustainability pressure, labour constraints, ESG expectations, brittle data, and the ongoing fantasy that more automation automatically means more resilience. Simon makes a crucial distinction: automation is not autonomy. AI agents can fetch missing data, open tickets, draft supplier communications, or propose replans, but they struggle when decisions require unresolved trade-offs between service, cost, cash, carbon, and customer risk.
What changed my thinking was the clarity of the failure modes. LLMs are brilliant explainers, but they are not supply chain decision engines. More data helps until the world changes, and under real uncertainty machine learning can become confidently wrong. Simon also makes the case for smaller, well-governed models over perfect digital twins nobody trusts. Sometimes the best predictor is embarrassingly simple, like the length of a service request. Kismet: apparently, after all the algorithms have finished tap-dancing, counting words can still beat the clever machine. Humanity is safe for another Tuesday.
This is for senior supply chain, procurement, logistics, operations, manufacturing, sustainability, and risk leaders who need practical AI decisions, not expensive mythology.
If you’re applying AI to real supply chain decisions, I’d like to hear where you draw the line between automation and accountability.
🔗 Podcast website: resilientsupplychainpodcast.com
🎧 Subscribe or follow for more conversations on supply chain resilience, operational risk, sustainability, and decision-making.
⏱️ Chapters / Timestamps
00:00 – AI Can’t Own Supply Chain Consequences
01:15 – Turning AI Hype Into Supply Chain Metrics
02:49 – Why AI Fails Without Operational Objectives
04:08 – Agentic AI, Autonomy, and Fast Mistakes
07:52 – Where AI Breaks: Cost, Carbon, and Service Trade-Offs
11:32 – Why LLMs Are Not Decision Engines
19:08 – When More Data Makes Forecasting Worse
21:02 – Simulation When Disruption Breaks Forecasts
24:13 – Supply Chain Resilience Beyond Forecast Precision
25:39 – Why AI Amplifies Human Expertise and Bad Assumptions
28:23 – The Risk of Overbuilt Digital Twins
37:59 – AI Governance Before Live Operations Break
