Agentic AI workflows
Agentic AI Workflows — From Prototypes to a Standard Capability Across Aspen Products
______________________________________________________________________________________
Case Study Snapshot
Context: AspenTech users operate in high-stakes industrial workflows (simulation/modeling + sensor-driven ops) where speed, accuracy, and confidence matter; the leap was multi-step automation that can plan and assist, not “more features.”
Problem: Core workflows were manual and expert-dependent, driving low time-to-value, variable outcomes, adoption friction, ticket volume, and heavy training burden.
What I led: Standardized agentic workflows into a repeatable portfolio capability (shared agent architecture, prompt templates, and reusable “agentic UI” interaction patterns) rather than one-off builds.
What shipped: Three flagship workflows:
Diagnose convergence failures + propose fix steps
Generate simulation setup from a goal + constraints
Auto-build dashboards/reports from selected KPIs
Trust & safety guardrails: human-in-the-loop approvals, audit trails + rollback, permissions gating, citations/confidence signals, and hallucination risk reduction via deterministic tools + restricted action spaces.
Outcomes: 50% time savings on targeted workflows, 30% reduction in support tickets, 80% adoption by teams, 90% reduction in training time, and double-digit renewal growth tied to customer value from the capability.
______________________________________________________________________________________
AspenTech’s software supports high-stakes industrial decision-making—simulation, modeling, and sensor-driven operations—where users need speed, accuracy, and confidence. The next leap wasn’t “more features,” it was automation that can plan, act, and assist across multi-step workflows.
Problem:
Core workflows were manual and expert-dependent: long sequences of steps, back-and-forth troubleshooting, and inconsistent outcomes. That created:
Low time-to-value (especially for new users)
Variability in results (dependent on senior expertise)
Ticket volume and adoption friction
Heavy training burden that limited scale
Approach:
I mobilized a global, cross-functional organization to scale agentic workflows from isolated prototypes into a repeatable, standardized capability across major Aspen products—starting with the simulators for modelers and process engineers, then expanding into sensor-leveraged products.
The key was standardization, not one-off builds:
Shared agent architecture
Repeatable UX patterns for agentic interactions
Standardized prompt patterns
Reusable system prompts + templates
“Agentic UI” component patterns and guidelines so teams didn’t reinvent the wheel
What shipped:
We productized three flagship agentic workflows that removed the manual grind and made expert behavior repeatable:
Diagnose convergence failures + propose fix steps
Before: manual troubleshooting across screens, tribal knowledge, and senior guidance
After: guided diagnosis with recommended actions, executed with confirmation (and safe auto-execution where appropriate)
Generate a simulation setup from a goal + constraints
Before: time-consuming configuration and iterative rework
After: intent-to-model setup that outputs a runnable simulation, with user confirmation at decision points
Auto-build dashboards / reports from selected KPIs
Before: manual dashboard creation and repetitive reporting cycles
After: automated dashboard/report generation driven by KPI selection and context
These patterns were delivered both in-product (end-user value) and through internal tooling agents (to accelerate delivery and consistency).
Guardrails:
Trust and safety were treated as product requirements:
Human-in-the-loop approvals for consequential actions
Audit trails and rollback for reversibility
Citations/sources, confidence signals, and transparent reasoning
Permissions and role-based action gating
Hallucination risk managed through deterministic tools and restricted action spaces
Built to meet expectations in regulated industries
Outcomes:
50% time savings on targeted workflows
30% reduction in support tickets
80% adoption by teams (standard capability used across product teams, not isolated pilots)
90% reduction in training time
double-digit renewal growth tied to customer value from the capability
Why it mattered:
This operationalized agentic AI as a true differentiator: not a demo feature, but a portfolio-level capability that transformed static tools into guided, adaptive systems—with guardrails that make automation trustworthy in high-impact, regulated environments. It also proved a scalable model: prototype → standardize → roll out across the portfolio in ~6 months, without every team inventing their own approach.