01
Capture
Inputs from email, notes, articles, meetings, or chat.
Agentic operating model / client-safe explainer
Agents are not the team. They are the operating layer.
A real-world look at how The Hooman Loop uses Hermes, a Mac mini, Obsidian, internal tools, and specialized agents to run strategy and design work with more continuity — without turning AI into a private shortcut.
The tension
AI can make individuals faster and teams less coherent.
The most important design problem is not whether AI can produce more output. It is whether a team can preserve shared context, review, trust, and strategic clarity when every person has access to more autonomous capacity.
Operating model
From prompt work to workflow design.
The system is designed around explicit movement of work, not one-off chatbot sessions.
02
Route
Work is assigned to the right context, report, or agent role.
03
Research
Agents gather public context, sources, and background.
04
Synthesize
Messy inputs become a structured point of view.
05
Draft
Briefs, outlines, reports, and messages take shape.
06
Review
Human judgment, quality gates, and source checks.
07
Act
Publish, send, schedule, decide, or hand off.
System stack
The stack is practical, not magical.
An always-on local machine, an agent runtime, durable memory, messaging surfaces, and named specialist roles.
Runtime
Hermes on a Mac mini
A local/private execution layer for tools, scheduled routines, research, file operations, and agent coordination.
Memory
Obsidian as source of truth
Human-readable notes, reports, decisions, and working artifacts keep agent work visible and reusable.
Surfaces
Telegram + internal tools
Fast capture in chat; structured queues, reports, landing pages, and operating views in private tooling.
What agents handle
Capacity work.
- Context recall across messy inputs
- First-pass research and synthesis
- Drafting alternate framings
- Routing and decomposition
- QA checklists and consistency checks
- Monitoring open loops
What humans keep
Judgment work.
- Strategic direction
- Taste and editorial standards
- Client relationship nuance
- Ethical boundaries and disclosure
- Prioritization and trade-offs
- Final approval
Anti-silo principles
Design the collaboration layer, not just the prompts.
Shared artifacts
Agent outputs should land in briefs, notes, reports, tickets, or decks — not disappear into private chat logs.
Visible state
Queues, statuses, and open loops make delegated work inspectable by humans and teams.
Named roles
Scout, Sterling, StraatBot, and specialists have explicit jobs so the system stays legible.
Review gates
High-stakes actions require human approval. Quality comes from the review architecture, not model cleverness.
Human-readable memory
The source of truth should be readable without an AI agent present.
Example workflows
Real work, not demo theater.
01 / Talk prep
Messy email → brief → deck → leave-behind
A client prompt becomes a structured talk thesis, slide arc, system map, and public-facing follow-up artifact.
02 / Scout research
People and organizations become durable intelligence
Reports, entities, and insights preserve context for strategy, relationships, and follow-up.
03 / Read-later memory
Articles and videos become project-ready notes
Inputs are saved, summarized, routed, and connected to active work instead of becoming a forgotten queue.
Takeaway
The future of design teams is not fewer humans. It is clearer human judgment surrounded by better operating layers.
The Hooman Loop
Design a team-safe AI operating layer.
For product, strategy, and consulting teams exploring how agents can create leverage without creating hidden process, incoherent outputs, or procurement anxiety.