Aydar Akchurin, a tribology researcher and creator of TriboSolver, has published a production agent workflow on TriboNet that uses OpenClaw to execute industrial contact simulations entirely through Telegram chat commands.
How the Pipeline Works
The workflow chains four stages: a Telegram request initiates the simulation, OpenClaw controls TriboSolver execution across the backend services, result artifacts are generated, a Hermes AI agent runs verification on the outputs, and a human performs the final check. The full pipeline handles multi-hour simulation runs for contact mechanics problems in industrial tribology, the study of friction, wear, and lubrication in mechanical systems.
“I communicate with this agent like with another human and I can set up the problem in this way, easily,” Akchurin wrote describing the setup.
Why This Matters Beyond Software
Most production agent deployments documented publicly sit in two verticals: software development (coding agents, CI/CD automation) and financial services (trading, compliance, back-office operations). This workflow puts agents into industrial engineering, where the workloads are computationally heavy, run for hours, and require domain-specific verification before results can be trusted.
The human-in-the-loop design is notable. Rather than replacing the engineer, the pipeline uses AI agents at two points: OpenClaw for execution orchestration and Hermes for automated result verification. The human retains final authority. That pattern, where agents handle the execution and preliminary verification while humans make the accept/reject decision, mirrors the governance frameworks that enterprise buyers have been demanding for agent deployments in regulated industries.
The Cross-Vertical Signal
Agent frameworks marketing themselves to developers and enterprises tend to showcase familiar use cases: customer support, code generation, data analysis. Industrial engineering workflows like tribology simulation require agents to manage long-running processes, coordinate with specialized scientific software, and produce artifacts that need domain-expert validation. The fact that a solo researcher can wire this up with off-the-shelf agent frameworks (OpenClaw plus Hermes) on commodity infrastructure suggests the tooling has matured enough for adoption patterns that vendor marketing hasn’t caught up to yet.