Agent workbench

Practical tools to test an operational workflow before giving an agent authority.

A small set of practical tools for operations teams: explore who owns each decision with the operations map, assess workflow boundaries, data readiness, failure visibility, and human review with the readiness canvas, and inspect the methods and source behind the public cases. Repeat the loop that matters — inspect → act → verify → hand over. The prototypes are educational and are not production automation.

My rule: a polished answer is not proof. A working page, repository, test result, or deployment link is much more interesting.

Try the public prototypes

Operations intelligence map

A decision map explored through six lenses: SCOR process, operating area, roles, systems, data, and AI role. Change the lens instead of pretending one diagram explains the whole operating system.

Explore the map

Agent readiness canvas

A small interactive HTML tool for checking workflow boundaries, data readiness, failure visibility, human review, and traceability before calling a use case agent-ready.

Open the prototype

One-page operations case library

A library of searchable analytical briefs across source, transform, plan, and fulfill. The methods, data requirements, and KPIs now live on one page; the original slide versions remain available.

Open all cases on one page · Inspect the source

Reusable practices

Evidence before enthusiasm

A curated source sheet for agentic AI in operations. Claims should lead to inspectable evidence, not a synthetic confidence trick.

Hermes Agent field guide

A practical reference for tools, skills, scheduled work, and verification. The interesting part is not prompting. It is dependable execution.

Inspect → act → verify → hand over

The pattern works for code, content, data, and operations: inspect the real source, make the smallest useful change, test it, then leave a crisp review path.

Keep the decision owner visible

An agent may prepare a route, risk signal, maintenance recommendation, or publication. A human still owns the exception, commitment, and reputation risk.

Inspectable work

Code and content should survive curious people clicking on them.

The public GitHub layer includes operational use cases, supply-chain analytics, SDG analytics, analytics translation, and this homepage. Some repositories are mature learning artifacts; some are experiments. They are labelled rather than dressed up.

View curated GitHub projects Open the GitHub account

What earns a place in the workbench

The workbench is for artifacts that turn a difficult operational question into something people can examine: a clear model, an interactive decision aid, or a bounded agent workflow with visible assumptions, failure paths, and review points. Each artifact should stand on its own, use public or synthetic material, and be reproducible without access to private infrastructure.

That standard is deliberate. No private data, hidden credentials, or employer-specific material — and no demo that depends on narration to conceal its limits. If another person cannot inspect the evidence, understand the boundary, and repeat the result, it is not ready to publish here.