BLUN Agents

AI roles that do the work, not just describe it

BLUN Agents are self-directed AI roles for recurring, multi-step work. An agent gets more than a name and a personality: it gets a concrete goal, a defined remit, the right BLUN model, a set of permitted tools and binding limits.

That turns an open-ended chat into accountable work. The agent knows what it is responsible for, which information it may use, which steps it can take on its own, and when a person has to decide.

You can start an agent for a single job or set one up permanently for a defined area. They suit any work that follows a recognisable pattern but still calls for judgement, decisions and the ability to adapt.

A role with real responsibility

Every agent is set up from the same clear parts:

• Role: What job does this agent do?
• Goal: What result should it reach?
• Remit: Which work falls to it?
• Model: How much capability does it need?
• Tools: Which systems and functions may it reach?
• Limits: Which actions are off-limits or need sign-off?
• Inputs: Which data and sources may it use?
• Result: In what form must the work be delivered?
• Check: How do you know the job was finished properly?

This structure keeps an agent from wandering outside its remit. Tasks, permissions and expected results all stay traceable.

From assignment to a result you can check

A BLUN Agent can work through a task in stages:

1. Understand the assignment: Goal, material and limits are reviewed.
2. Plan the approach: The agent breaks the task into sensible steps.
3. Choose the tools: Only permitted functions and data sources are used.
4. Do the work: Information is processed and actions are carried out.
5. Check the interim results: Errors and missing details are caught.
6. Escalate where needed: Unclear or restricted decisions are handed on.
7. Verify the result: The output is measured against the goal and the acceptance criteria.
8. Report completion: The result, the steps taken and anything still open are summarised openly.

An agent is therefore not finished simply because it produced a convincing-sounding answer. The outcome has to match the agreed goal and the stated acceptance criteria.

The right model for each role

BLUN Agents can draw on different models depending on the task:

• King for long-running, complex and tightly interdependent work
• Queen for professional day-to-day work, development, analysis and dependable execution
• Prince for fast, clearly defined and frequently repeated tasks
• Voice for spoken input, telephony and voice-driven agents
• Imagine for visual work and image production

An agent can stay on one model, or hand parts of a job to a more suitable tier within a controlled flow. That way the heaviest capability is only used where it actually earns its keep.

Tools instead of isolated answers

An agent becomes genuinely useful once it can work with real systems rather than only produce text. Through defined tools and the BLUN MCP Server, agents can reach files, databases, APIs and internal business functions.

Agents can, for example:

• retrieve information from a system
• read and structure documents
• prepare tasks or tickets
• check and update data
• produce reports
• analyse and edit code
• write content for websites or campaigns
• keep an eye on recurring processes
• hand results on to the next step

What an agent is actually allowed to do is set by its configuration and by the permissions on each tool.

Several agents as a team

Complex work can be split across specialised roles: one agent plans and coordinates, another does the specialist work, a third checks the quality.

Clear roles prevent duplicated effort and conflicting changes. Each agent gets its own bounded remit, and results can pass between roles without losing sight of the shared goal.

Team setups might look like:

• product lead, development and quality assurance
• research, editing and sign-off
• support, technical analysis and escalation
• campaign planning, copy and image production
• data checking, reporting and an executive summary

Control, limits and human sign-off

Autonomy does not mean unlimited access. BLUN Agents are meant to work within clear permissions, and critical actions can be blocked outright or made conditional on confirmation.

Common limits include:

• read-only access to sensitive data
• changes confined to one specific project
• no publishing without sign-off
• no messages to outside recipients without confirmation
• a restricted set of tools and data sources
• defined cost or run-time ceilings
• handover to a person whenever something is unclear

That lets you automate work without quietly handing responsibility and control over to a model.

Common agent roles

• support agent for recurring customer enquiries
• development agent for well-defined software tasks
• QA agent for checks and deviations
• research agent for structured information gathering
• editorial agent for content and publishing
• project agent for tasks, risks and progress
• operations agent for internal processes
• data agent for validation, extraction and reporting
• voice agent for spoken interaction

When BLUN Agents are the right choice

Use BLUN Agents when:

• a job takes several steps,
• work recurs along similar lines,
• tools and data sources have to be used,
• roles and responsibilities need clear boundaries,
• a result has to be verified, not merely written up,
• several specialised AI roles should work together,
• human sign-off must remain in place at the critical points.