Direct answer – What changed with Eyelit 10.1 Agent EyeQ?
Eyelit 10.1 moves Agent EyeQ beyond answering planning questions toward logged execution across MES, APS and SIOP operations. Caddy mode keeps the user in charge; Employee mode can own bounded work and escalate exceptions. Eyelit describes approval gates, read/write separation and replayable actions, but has not published independent 10.1 customer-performance benchmarks.
Eyelit announced Version 10.1 in July 2026 with Agent EyeQ positioned to execute planning and scheduling decisions, rather than stop at decision support.
The release spans more than the agent. It adds transparent demand forecasts with confidence intervals, lifecycle-aware demand rules, new APS constraints and schedule controls, component-level MES processing, expanded APIs and FDA-oriented label and history-record functions.
The sharp buyer question is not whether the system uses agentic AI. It is which actions the agent can take, when a person must approve them, how a manual choice survives the next reschedule and whether the audit trail is enough to reconstruct a consequential decision.
Key Takeaways
- Eyelit says Agent EyeQ can call more than 1,700 operations across MES, APS and SIOP.
- Caddy mode suggests and executes on request; Employee mode can operate against assigned goals and escalate exceptions.
- Version 10.1 adds persistent Gantt locks, AI infeasibility explanations and live schedule response to estimated downtime.
- Eyelit says actions are logged, traceable, approved and replayable with human-in-the-loop controls.
- The release provides no independent benchmark for schedule quality, planner time saved or autonomous error rate.
What Eyelit 10.1 released
Eyelit describes Version 10.1 as an integrated manufacturing operations management release for complex discrete manufacturers. Demand Planning gets role-specific workspaces, model selection and blending, probabilistic forecasts with confidence intervals, lifecycle events and priorities that can put safety-stock replenishment ahead of less critical forecast demand.
The 10.1 release notes also expand Advanced Planning and Scheduling. New functions cover oven and autoclave batches, shared resource pools, machine and cross-operation affinities, sales-order pegging and automatic schedule alignment when expected asset downtime exceeds a configurable threshold.
MES changes include targeted processing for selected components without splitting a lot, operation previews, advanced label-printing controls and expanded electronic component history records. That breadth raises integration and validation cost while making the record layer more important. Our manufacturing software stack map explains why execution and planning cannot both own the same decision silently.
What Agent EyeQ can change in a live schedule
Eyelit says Agent EyeQ can discover and call more than 1,700 platform operations across MES, APS and SIOP. The advertised workflow can plan, release, dispatch, hold and schedule work end to end.
Version 10.1 gives that agent a more controllable scheduling surface. A manual drag-and-drop move in the Gantt chart can persist as a resource lock through later reschedules. When the solver finds no feasible schedule, an AI agent can generate a natural-language explanation of conflicting constraints. Estimated downtime can trigger an automatic schedule update instead of waiting for a planner.
The boundary with APS still matters. The CAI PlanetTogether acquisition framed APS as the layer that builds a feasible schedule between ERP and MES. Agent EyeQ adds a decision-execution loop around that layer, but it does not remove the need for accurate routings, calendars, materials and work-center constraints.
Where human approval and audit controls remain
Eyelit presents two modes using the same tools and governance. Caddy mode works beside a planner or operator, suggesting, drafting and executing when asked. Employee mode receives a goal, runs within assigned constraints and escalates exceptions to a human manager. The product page describes human-in-the-loop approval gates, read/write separation and notifications before changes reach production.
Every action is described as logged and replayable, with approved runs feeding a quality-gated knowledge base. Eyelit also says the reasoning layer is model-agnostic and tenant boundaries protect customer data. Buyers still need the configured approval, failure and rollback rules.
Our read: the most valuable 10.1 feature may be the ability to preserve intent, not the word autonomous. Persistent locks, traceable actions and explanations for infeasibility give planners a way to see why the schedule changed. That is the test a MES or MOM evaluation should apply before counting an agent as production control.
What manufacturers should test before Employee mode
Begin in Caddy mode on one bounded scheduling job. Use a work center where due dates, changeovers, material constraints and downtime are already measured. Compare schedule feasibility, planner touches, late orders and manual overrides against the existing process for several cycles.
Then force failure conditions. Remove a material, extend a machine outage, introduce a rush order and give the system a conflict it cannot solve. Confirm that the agent stops at the correct approval gate, explains the infeasibility, preserves authorized manual locks and leaves a complete record. The resilience principle behind plant-edge execution applies here too: the production plan needs a safe state when connectivity or a service is unavailable.
Inspect permissions at the operation level. A planner who can simulate a schedule should not automatically be able to release lots, hold work or change a production record. Require named owners for approval policies, model changes, knowledge-base updates and rollback. Replay one decision from input through tool calls to final action before approving broader autonomy.
Finally, demand a benchmark that matches the purchase claim. Eyelit has published detailed capabilities, but the 10.1 materials do not provide customer results for planner hours, on-time delivery, schedule stability or agent error rate. Price the integration and validation work as part of the program, not only the licence.
Frequently Asked Questions
Agent EyeQ is Eyelit’s agentic AI engine for manufacturing operations. Eyelit says it can reason over live platform data and call more than 1,700 MES, APS and SIOP operations to plan, release, dispatch, hold and schedule work, with approvals and an audit trail.
Version 10.1 adds Agent EyeQ execution, redesigned demand planning, explainable forecasts, lifecycle rules, safety-stock priorities and deeper APS controls. Scheduling additions include batch resources, affinities, persistent Gantt locks, infeasibility explanations, sales-order pegging and automatic reaction to estimated shop-floor downtime.
Eyelit says yes, when configured for Employee mode and operating within assigned goals, constraints and human-in-the-loop guardrails. Caddy mode keeps the user in charge and executes on request. Buyers should verify the exact approval policy, permissions, rollback and safe-failure behavior in their own configuration.
Caddy mode assists a person by suggesting, drafting and acting when requested. Employee mode owns a defined job, works toward assigned KPIs and escalates exceptions. Eyelit says both modes use the same tools and governance, but the operating configuration determines where approval gates are placed.