AI Growth Department

Mission Growth's decision logic is run by a structure of specialized agents, not a single model. The structure does three things:

1. Reaches the data: Collects the necessary signals.
2. Builds context: Interprets data with domain rules and past learnings.
3. Turns it into action: Converts the result into actionable recommendations.

Hierarchy

Kayra runs as the top-level orchestrator. Kayra does not try to solve every problem alone; the work is dispatched to the right specialist, the outputs are combined, and the final answer is presented to the user. This setup produces more disciplined and more consistent decisions across different areas.

Specialization areas

Each agent specializes in its own decision area. The CRO agent focuses on conversion flows, the SEO agent on organic visibility, the RevOps agent on sales and marketing alignment. Kayra's job is to assemble the right combination while respecting these specialization boundaries.

How agents operate

Each specialist agent runs under a fixed discipline. This discipline generally includes:

• A clear role and area of responsibility
• An interpretation framework specific to its domain
• A consistent output format
• Explicit autonomy and approval boundaries

This setup lets different agents exhibit different specialist behavior even when the same base model family is used. The difference sits at the level of decision approach, not just tone.

Tools, skills, and automations

Agents work across three layers.

Tools reach external systems in a single action. Pull Google Ads campaign performance, fetch a GA event list, read a customer record from the CRM. One request, one response.

Skills are the agent's "this is how we do this work" playbook. When the SEO agent runs a landing page audit, it reads from a skill which order to inspect things in and which checks to perform. The skill guides interpretation; it is not directly executed code, and the agent decides on its own how to use a tool.

Automations are multi-step chains where one agent's output feeds the next agent's input. They run in two forms:

• Built-in automations: The library holds over 320 templates such as competitor-strategy, daily memory reflection, and goal deadline reminder. They start on a trigger or a schedule.
• On-the-fly automations: When a request without a ready template arrives, Kayra designs the automation itself: which step, which agent, in which order. The plan is built at runtime, with a 15-step ceiling.

The three layers work together. A tool's output becomes an input to an automation, and a skill's interpretation steers that automation's step.

Call chain example

The user asks: "Why did CPL on Meta rise this week?"

1. Kayra reads the question and classifies the area as PPC.
2. The PPC agent collects the relevant performance data.
3. Where needed, competitor moves, similar past situations, and current campaign context are evaluated together.
4. The agent produces the output under three headings:
   • Finding: The critical observation drawn from the data
   • Action: Steps to take this week
   • Context: The background the conclusion rests on
5. Kayra collects the output, evaluates its quality, trims unnecessary detail, and presents the answer to the user. At the same time, the learning is written into the space memory; when the same question comes up again, the system starts by recalling the prior decision.

Throughout this process, the system surfaces both the answer and the context the decision is built on.