— Quick Read

The Orchestrator Agent: Connecting Multiple Agents for Interview Prep

AI agents become more useful to a business when they are treated as reusable building blocks, not one-off chat sessions.

This page explains a real orchestration pattern: one agent accepts a company and interviewee, sends research work to three specialist agents, then combines the results into a business-ready interview prep report for the Future State of Streaming series.

Want the full recipe? Here is the reusable guide an AI builder can follow.

You can also try the finished agent for yourself here.

What You Are Building

You are building an interview-prep workflow that starts with four simple inputs:

  • Company name
  • Company website
  • Interviewee name
  • Interviewee LinkedIn URL

The workflow then produces a markdown background report a business team can use before an interview, sales conversation, executive briefing, podcast recording, webinar, or content-planning session.

The key idea is orchestration. The main agent does not try to research everything itself. It delegates specialized work to other agents, waits for those research jobs to finish, and then uses an AI synthesis stage to turn the separate inputs into one coherent report.

What the Tools Do

  • Blocks is the agent network. It lets one agent call another agent by name, so a workflow can be assembled from specialist capabilities instead of rebuilt every time.
  • The FSOS Interview Prep Orchestrator is the coordinating agent. It receives the business request, starts the research tasks, gathers the results, and creates the final report.
  • Railway is where the orchestrator runs continuously online. Instead of starting the agent manually each time, Railway keeps the provider worker available for Blocks requests.
  • OpenAI is the language model service used for final synthesis. It takes the research packet and the FSOS one-pager, then drafts a readable, structured prep report.
  • Apify is used by the LinkedIn research agents to collect structured information from public LinkedIn pages and posts.
  • GitHub is where the reusable instruction file lives, so another AI builder can reproduce, update, or adapt the workflow.
  • An API key or token is a private password-like credential that lets the agent use a service such as Blocks, OpenAI, or Apify.
  • A CLI is a command-line tool the builder uses to register, check, or deploy the agent.
  • A dataset or artifact is the structured output returned by a research run, such as a markdown report, JSON evidence file, or source summary.

The Sub-Agents It Calls

The orchestrator currently calls three Blocks agents in parallel.

Company Profile Research Agent

Agent name: company_profile_research_agent

This agent researches the company itself. It builds a sourced company profile covering what the business does, its products, customers, market position, leadership signals, competitors, risks, and recent developments.

Business value: it gives the report a factual company foundation before the workflow starts interpreting messaging or preparing questions.

LinkedIn Company Report Agent

Agent name: linkedin_company_report_agent

This agent looks at the company’s public LinkedIn activity. It identifies messaging themes, notable posts, content patterns, and current market-facing signals.

Business value: it shows what the company is emphasizing right now, which can reveal campaigns, events, partnerships, hiring themes, product pushes, or positioning shifts.

LinkedIn Person Report Agent

Agent name: linkedin_person_report_agent

This agent researches the interviewee’s public LinkedIn profile and recent activity when available. It captures role context, professional background, public themes, and conversation starters.

Business value: it helps the team prepare for the person, not only the company. That makes the final brief more useful for interviews, outreach, and executive conversations.

What the Guide Gives You

The linked guide is for an AI builder or technical operator. It gives them the concrete steps to reproduce or maintain the workflow. It includes:

  • The required inputs and optional settings.
  • The current downstream agents and what each one receives.
  • The environment variables needed in Blocks, Railway, OpenAI, and Apify.
  • The validation steps for checking the agent locally.
  • The deployment steps for Railway.
  • A smoke-test request.
  • Troubleshooting guidance for crashed services, missing credentials, failed downstream agents, and synthesis fallback.

For a business reader, the important point is this: the skill file turns the workflow from tribal knowledge into reusable operating instructions.

Before You Start

  • Blocks account, so the orchestrator can be registered and called by other agents or workflows.
  • Railway account, so the orchestrator can run continuously online.
  • OpenAI API key with billing enabled, so the final synthesis stage can turn the separate research results into a polished report.
  • Apify account and API token, because the LinkedIn sub-agents use Apify to collect structured public LinkedIn activity.
  • Access to the three sub-agents named above, because the orchestrator depends on them to complete the company, company LinkedIn, and person research steps.
  • The FSOS one-pager, because it gives the workflow a consistent strategic lens instead of producing a generic company brief.
  • The company name, company URL, interviewee name, and LinkedIn profile URL for each report.

If someone else is setting this up for you, the main credentials they will ask for are BLOCKS_API_KEY, OPENAI_API_KEY, and APIFY_API_TOKEN. Treat these as private credentials.

How It Comes Together

The user submits the company and interviewee details. The orchestrator then starts three research jobs at the same time:

  • Company profile research
  • LinkedIn company activity research
  • LinkedIn person research

Running the agents concurrently matters because the workflow does not wait for one research job before starting the next. That makes the report faster than a step-by-step manual process.

When the research agents finish, the orchestrator collects their markdown and JSON artifacts. It also reads the FSOS one-pager, which contains the point of view and topic pillars for the Future State of Streaming series.

The AI synthesis stage then combines everything into a markdown report designed for business preparation, not raw technical inspection. The final output usually includes:

  • Executive brief
  • Interview target summary
  • Company background
  • Interviewee snapshot
  • Why the guest fits Future State of Streaming
  • Strongest FSOS alignment
  • Interview prep angles
  • Suggested questions
  • Content generation hooks
  • Data notes

The system is also modular. If a faster or better company research agent becomes available, that sub-agent can be swapped without rebuilding the whole orchestrator. This already happened in this workflow: a slower deep research agent was replaced with company_profile_research_agent, while the rest of the workflow stayed intact.

Starter Prompt

Use this skill file:
https://raw.githubusercontent.com/CagedEther/agentic-business-schooled/refs/heads/main/fsos-interview-prep-orchestrator-skill.md

Build or update a Blocks orchestrator agent for interview prep. The agent should accept companyName, companyUrl, personName, and linkedinUrl. It should call company_profile_research_agent, linkedin_company_report_agent, and linkedin_person_report_agent concurrently, then synthesize the results with the local FSOS one-pager into a markdown background report.

Target company: {{company_name}}
Company URL: {{company_url}}
Interviewee: {{person_name}}
LinkedIn URL: {{linkedin_url}}

Keep the output business-readable, include data notes, and do not expose API keys or private credentials.

What a Good Result Looks Like

A good result feels like a brief a sharp producer, marketer, salesperson, or executive staffer would actually use.

It should be specific to the company and person. It should connect research signals to the FSOS point of view. It should include useful interview angles and questions, not just a summary of facts.

It should also be honest about uncertainty. If LinkedIn activity is sparse, the report should say so. If a downstream source fails, the report should preserve that as a data note rather than pretending the research was complete.

The best version gives the business team three things:

  • Context: what to know before the conversation.
  • Relevance: why this company or person fits the FSOS series.
  • Action: what to ask, publish, or follow up on.

Use It Responsibly

Treat the output as a preparation brief, not a final fact-checked dossier. Review important claims before using them in public content, sales materials, or executive communications.

Use LinkedIn research only for public, business-relevant context. Do not use it for sensitive personal profiling or private-data inference.

Watch API costs. LinkedIn comments and reactions are intentionally disabled by default because they can increase Apify usage and make the output noisier.

Protect credentials. API keys and tokens should never appear in screenshots, published docs, commits, or shared reports.

Other Ways to Configure These Agents

The same sub-agents can be recombined for other business workflows.

Sales account prep: company profile research + LinkedIn company activity + optional person research for the buyer or executive sponsor → account brief, outreach angles, discovery questions, and likely priorities.

Executive briefing prep: company profile research + LinkedIn person research + optional internal account history → executive memo, relationship context, key risks, and talking points.

Content strategy research: LinkedIn company activity + company profile research + optional competitor research → messaging themes, content gaps, campaign angles, and thought leadership opportunities.

Partnership research: company profile research + LinkedIn company activity + optional partner-category or market-map agent → partner fit analysis, joint messaging ideas, integration opportunities, and open questions.

Event speaker prep: LinkedIn person research + company profile research + LinkedIn company activity → speaker brief, panel questions, topic alignment, and promotional content hooks.

Competitive messaging monitor: LinkedIn company activity for several companies + company profile research for each + optional synthesis agent that compares positioning → competitive messaging report, market narrative shifts, and differentiation opportunities.