We stand at a watershed in enterprise planning activities as a result of the availability of artificial intelligence (AI) and agentic systems. In particular, AI and agents will facilitate the process of continuous supply chain planning through better cross-functional coordination, significantly shortening the time it takes to iteratively juggle and resolve supply and demand chain objectives and constraints across multiple time horizons. They will significantly reduce the manual steps of supply and demand planning under dynamic business conditions, which occupy much of the time of those involved in planning and execution management.
AI and agents have the potential to substantially shorten planning and execution cycle times to make enterprises more agile and efficient. Rather than having to focus almost all their
Supply chain planning providers have been quick to recognize the need to embed AI and agentic systems in their software, but enterprises should regard this as an evolving process. A realistic evaluation checklist at this point would have a “sort of” column in addition to the standard yes/no options. Nonetheless, there has been significant progress over the past two years in the following ways:
AI-enabled supply chain planning is evolving from simple copilots to more useful agents. Generative AI initially focused primarily on answering questions and explaining forecasts, but increasingly it can identify issues, reason about causes, propose actions and act autonomously within guardrails. From the provider’s standpoint, buyers should confirm that claimed capabilities are available. They also must assess their readiness in terms of data quality and accessibility, operating model and decision logic alignment with the software, process design, integration with all necessary systems (including homegrown) as well as governance. All of these are necessary to ensure that the surrounding environment allows available agents to make useful, trusted and executable decisions.
Agentic planning software extends planning functionality into execute mode. This has the potential to meaningfully shrink decision latency and allow a faster coordinated response. Whether and in what way enterprises integrate these extended capabilities with their supply chain management application (or applications) is an architectural issue that is never one-size-fits-all.
Multi-agent orchestration is becoming the architectural norm because it can improve planning performance using specialized agents optimized for different domains (such as demand, supply, inventory and finance), while reducing friction in coordinating their recommendations against unified enterprise business objectives. Ideally, this can improve decision quality, reduce response times and allow organizations to evaluate consequences simultaneously rather than through sequential planning processes. It also supports a more loosely coupled approach to architecture. Agents can be introduced, tested and refined incrementally, and they can be less costly to operate because compute requirements can be optimized for each agent. While useful, orchestration can add complexity that leads to conflicting objectives and suboptimal design unless priorities, constraints and decision rights are clearly defined and aligned with enterprise-wide objectives. Decisions must be explainable through observability and traceability. Providers must demonstrate how their agents share context, resolve conflicts and coordinate actions. In boardroom reviews, they must be able to show how agent orchestration deals with conflicting objectives, handles exceptions and reacts when it encounters bad data.
Agent studios are now table stakes, enabling individual enterprises to effectively configure or create agents that align with their business systems or niche vertical requirements rather than relying exclusively on what’s in the box. Agent studios also enable software providers to host vetted marketplace solutions created by implementation or service provider partners to expand the utility of their core system.
Governance and human-in-the-loop capabilities are essential. ISG’s 2026 AI Value Study finds that more than half of enterprises rely on humans in the loop (HITL) rather than autonomous agents for a range of AI-enabled processes. For more complex or consequential processes such as decision support or fraud detection, more than two-thirds require a HITL. Software must offer dependable explainability, provide audit trails, manage permissions, respect policy constraints, feature multivariate approval thresholds and escalation parameters. Enterprises expect the ratio of autonomous agents to rise by the end of 2027, but it’s likely the degree of increase will be specific to each enterprise and process.
The growing use of AI and agentic capabilities in operations is creating an opportunity for supply chain organizations to reduce the time between identifying a business condition, making a decision and executing the appropriate response. Buyers should not treat reduced latency as inherently valuable but continually assess whether faster decision-making produces a measurable improvement in financial or operational performance.
Moreover, to take full advantage of AI systems and agents, enterprises will have to make the necessary changes to their operating environment and address IT issues to avoid costly mistakes. Agents are not a flip-the-switch capability. AI and agents present new operating model issues. Pure autonomy is a practical impossibility today, so enterprises will need to define when agents may recommend, when they may act, when humans must approve and how exceptions are escalated. Our research finds that only 13% of agentic processes will be fully autonomous through the end of 2027.
Agent design and controls represent more than technical configuration challenges. They require a business-focused strategy that defines governance, risk management, process redesign and change management policies that guide the foundation of an enterprise’s agentic systems. Providers may offer some necessary support for these targets, but identifying and achieving longer-term business model objectives will require a clear plan and process for transitioning toward them.
For example, agents may not address latency issues or improve performance without considering the full end-to-end series of steps involved. As a rule, fast and accurate analytics will not achieve objectives unless the full scope of approvals, organizational processes and execution systems are assessed and changed to achieve faster signal-to-action intervals. Latency must be measured consistently from data acquisition, exception detection, diagnosis, decision formulation, approval and execution. Ideally, all of these should be tuned to exploit the advantages that agents can provide.
Data management will be an important factor in achieving the full potential of AI and agents in supply chain planning. Because of the dispersed functional nodes necessary to support
Supply chain planning groups should evaluate the impact of AI and agentic systems on decision latency in the context of a portfolio of material business decisions rather than any generic measure such as improved planning speed. For each decision, organizations should quantify current latency, the economically useful response window, value at risk, frequency, reversibility and feasible level of autonomy. This approach separates three distinct sources of value—productivity, decision quality and decision velocity—to establish where reduced latency provides a meaningful, measurable business advantage rather than simply a technical capability.
I also strongly recommend buyers take a pragmatic approach that reasonably tests the limits of what’s achievable. With a skeptical eye, buyers should assess the maturity of specific agents, because the devil is in the details. Test use cases against measurable outcomes such as reductions in cycle time, inventory turnover, fulfillment rates, exception rates and manual effort. Look for forecast improvement, reduced transportation costs or service-level improvement. Examine governance capabilities, auditability, integration depth, human-in-the-loop controls and lifecycle management. Their focus should be on overall productivity, including what they can do now that they could not do before, not just efficiency.
Regards,
Robert Kugel