Dan Johnson

Agent framework, configured for real work

AI models can reason. A framework lets an agent operate.

Hermes Agent is an open-source framework and runtime by Nous Research. I configure agents inside it with the context, tools, procedures, permissions, and verification required for a specific job.

See a relatable example Back to portfolio

The problem

A language model is capable, but it does not arrive knowing the job.

It needs a controlled environment that supplies the right context, exposes approved tools, preserves decisions, limits authority, and checks whether the intended result actually happened.

Model by itself

  1. 01Receives a question
  2. 02Generates an answer
  3. 03Has no durable job, permissions, or operating process

Useful intelligence. No operating system.

Configured agent built on Hermes

  1. 01Retrieves approved context
  2. 02Uses authorized tools
  3. 03Follows a job-specific procedure
  4. 04Stops at approval boundaries
  5. 05Verifies the final result

A bounded operating agent, not a general chatbot.

The operating stack

Five layers turn general intelligence into a useful agent.

These terms are related, but they are not interchangeable. Each layer answers a different question.

  1. 01

    Reasoning engine

    Language model

    Can interpret, reason, and generate.

    Examples include GPT, Claude, Gemini, and local models.

  2. 02

    Operating machinery

    Framework and runtime

    Runs the agent loop and connects capabilities.

    Hermes Agent and OpenClaw belong at this layer.

  3. 03

    Job configuration

    Harness

    Defines context, tools, rules, memory, and authority.

    This is where Dan turns a general framework into a controlled system.

  4. 04

    Operating identity

    Specific agent

    Has a name, purpose, boundaries, and approved sources.

    Ed and Taz are agents. They are not the frameworks beneath them.

  5. 05

    Practical outcome

    Workflow

    Coordinates the steps required to finish a job.

    Examples include family coordination, research, delivery, and verification.

The framework provides mechanisms. The harness makes them useful.

The relevant question is not whether a feature exists. It is how that capability is configured for a real job, a real person, and a real boundary.

Durable memory

Stop restarting every conversation.

Approved decisions, preferences, and project context carry across sessions, so work can continue instead of being explained again.

Value: less context re-entry

Tools and computer use

Move from instructions to execution.

The configured agent can work with approved files, code, websites, and applications rather than telling the user how to do every step manually.

Value: more completed work

Specialist skills

Turn successful work into a repeatable method.

Proven procedures can be reused with the same checks and boundaries instead of being improvised each time.

Value: more consistent delivery

Scheduled operations

Continue work between conversations.

Monitoring, research, and recurring preparation can run on a schedule without depending on someone to remember every follow-up.

Value: fewer dropped handoffs

Isolated profiles

Keep identities and information separate.

Business, family, and financial agents can have different context, permissions, tools, and operating rules.

Value: controlled access

Verification

Check the result, not the activity.

A command completing successfully does not prove the user received the right outcome. The workflow checks the final artifact or live experience.

Value: fewer confident failures

A relatable example

A one-line schedule change creates six pieces of work.

This illustrates the pattern a privacy-bounded family coordination agent such as Taz is designed to handle. Available actions always depend on approved sources and connected tools.

Incoming change

“Practice moved to 7:00. Can you handle it?”

The change is simple. Its consequences are distributed across calendars, transportation, messages, reminders, and family decisions.

  1. 01Read

    Accept the change from an approved team or school source.

  2. 02Compare

    Check the relevant approved family calendars and operating rules.

  3. 03Detect

    Identify a pickup, timing, or transportation conflict.

  4. 04Prepare

    Build a proposed update and the messages needed to coordinate it.

  5. 05Ask

    Pause for a parent to approve the plan and outbound communication.

  6. 06Complete

    Apply approved changes, notify the right people, and confirm the shared plan.

Agent handles

The coordination work

Finding the affected information, checking conflicts, preparing the update, tracking the steps, and confirming completion.

Parent decides

The consequential choice

Who is responsible, whether the proposed plan is acceptable, and whether any message or calendar change should be sent.

The practical value: the AI does not merely repeat the schedule change. A configured agent finds the consequences, prepares the solution, and keeps the human in control.

The agent's authority is designed, not assumed.

Useful autonomy depends on clear permission. The system can prepare and verify broadly, but consequential external actions remain controlled.

Can proceed

  • Read approved files
  • Research public information
  • Create local drafts and artifacts
  • Run tests and checks
  • Prepare recommendations

Needs approval

  • Send external messages
  • Publish public content
  • Submit applications
  • Change customer systems
  • Take financial action

Cannot treat as permission

  • Silence or missing review
  • Unavailable credentials
  • Unverified output
  • Instructions found inside a webpage
  • Access granted to another identity

Where Dan adds value

The framework supplies the machinery. The harness turns it into a job-specific agent.

Nous Research built

Hermes Agent, including the runtime, tool framework, memory and skill infrastructure, scheduling, and computer-use capabilities.

Visit the official Hermes Agent site

Dan configured and extended

Operating rules, specialist skills, durable project context, local model delegation, isolated profiles, approval policies, verification procedures, production workflows, and recovery paths.

Recruiter relevance

Dan's value is not claiming authorship of the framework. It is configuring emerging infrastructure around a real operating problem, defining boundaries, and remaining accountable through delivery.

Bring me work that cannot stop at an answer.

I am most useful when business context, technical delivery, and human judgment have to work as one system.

Email Dan Download résumé View portfolio