In September 2026 the World Economic Forum and Accenture published an insight report on how a new generation of AI start-ups is changing supply-chain resilience. It draws on more than 50 interviews with industry leaders, start-ups, investors and academics, and an analysis of more than 600 AI start-up solutions. This article summarises it and sets out what it means for the people who have to govern the decisions those solutions take.
What the report says
The report defines resilience as the ability to sense disruption early, absorb its effects, adapt operations and recover quickly enough to maintain performance. Most organisations are far from that. Only 29% of supply-chain organisations say they have built the capabilities they need, only 7% can execute decisions in real time, and supply-chain disruption is estimated to cost up to $1.6 trillion in missed revenue growth a year.
Its three findings for leaders:
- Resilience is now a strategic asset, and AI lets organisations pursue resilience and efficiency at the same time.
- AI start-ups are speeding up access to frontier capabilities, but the ecosystem moves fast and needs continuous monitoring: the areas attracting the most investment are not always where agentic AI is advancing fastest.
- Building everything internally is becoming the exception. The greatest value comes from a disciplined choice between buying, partnering and building, with human oversight and governance built in from the outset.
The report describes three operating models for agentic AI: human-to-agent (a person initiates and decides every step), agent-to-human (the agent acts within defined boundaries and escalates at critical points) and agent-to-agent (networks of agents plan, delegate and execute whole workflows, with people governing). It places most supply-chain domains at about level 1 to 2 on a four-level autonomy scale now (zero to two years), and expects 3.5 or more over the next five years and beyond. That is the report's market scenario, not a measured result. Organisations with autonomous capabilities expect to cut disruption response and recovery times by 62% and 60%.
Procurement shows the strongest agentic momentum of the six domains studied, with the largest share of funded agentic start-ups (27.4%) and of new ones (36.6%). The report also gives a maturity journey from basic to leading: for basic organisations autonomous execution is out of reach, for digital ones it is premature (bounded pilots only, with tight guardrails and human oversight), and for advanced ones it is achievable in bounded, well-governed domains.
The Aidecis angle
The report's message to buyers is to source AI from several start-ups and partners rather than build it all. That is sound advice, and it has a consequence the report touches only briefly: a supply chain will soon have agents from many vendors acting in it, next to the company's own planning algorithms and its people. Each vendor ships its own audit trail and override controls, for its own agent only. Nobody owns the view across all of them.
That cross-vendor view is what Aidecis is built for: one authority model for who or what may decide each consequential decision, within which limits, and one decision record across every actor, linked to the outcome.
- The agent-to-human model needs boundaries and escalation points. In Aidecis these are decision envelopes per person, seat and agent, checked before an action runs.
- Not every vendor's agent will check an Aidecis grant before it acts. Where an agent writes in its own system, Aidecis applies observed control: the write is read back, matched against the agent's registered role and limits, and flagged within minutes if it went outside them. It does not prevent the action; it makes it visible and lets the client move that agent to enforced control.
- Procurement is where vendors' agents arrive first, so it is where governing a vendor's agent starts.
- The maturity journey caps how far autonomy should go. Aidecis records the client's maturity per decision family and never proposes the bot mode where the report calls autonomy premature or out of reach.
- Resilience is measured, not assumed: response time and recovery time per decision type, against the client's own baseline.
- Buying from a young vendor carries continuity risk. Aidecis answers it with a continuity package in every subscription: export in open formats, an open-source ledger verifier, escrow and exit assistance, with a dedicated control plane as an option.
The report ends by saying the leaders in resilience will be those willing to build capability ahead of full clarity. In our experience that only works if the authority to act is clear before the agents arrive.
See how it works, explore use cases, or discuss a pilot.
Contact