DEV Community

Morgan Atkins
Morgan Atkins

Posted on

How to Build Three AI Agents That Replace Your First Three Hires

#ai #machinelearning #productivity #startup

Inspired by Khairallah AL-Awady's thread on agentic workflows for solo founders, @eng_khairallah1 on X

Every solo founder hits the same ceiling. Revenue is coming in, but not enough to justify hiring. So you keep doing everything: marketing, research, customer support, operations, bookkeeping. You become the bottleneck for your own business.

The shift happening in 2026 is that the smartest early-stage founders aren't hiring their first employees: they're building them. Using Claude, MCP servers, and agentic workflows, you can stand up three AI agents in three weeks that collectively handle the work of a full research analyst, content producer, and chief of staff.

This post covers the technical architecture for all three.

The Foundation: MCP Servers

Before building any agent, you need the tool layer. Model Context Protocol (MCP) is an open standard that lets AI models connect to external systems: your email, calendar, databases, the web: through a unified interface.

Every agent in this architecture relies on MCP servers for its capabilities. Think of them as the arms and legs of each agent: the Claude model is the brain, the MCP servers are everything it can reach out and touch.

Core MCP servers you'll need across all three agents:

@modelcontextprotocol/server-filesystem: read/write local files, knowledge base
@modelcontextprotocol/server-google-drive: access docs, sheets, shared knowledge
@modelcontextprotocol/server-gmail: email read/triage/draft
brave-search MCP: web search and monitoring
Enter fullscreen mode Exit fullscreen mode

Install and configure these in your MCP host (Claude Desktop, Claude Code, or a custom harness). Each server exposes a set of tools that Claude can call as part of its reasoning.

Agent 1: The Research Agent

What It Does

Your research agent is a proactive market intelligence analyst. It doesn't wait to be asked: it monitors the landscape continuously and delivers structured weekly briefs before you start your week.

Most founders do research reactively. A research agent does it before you know you need it.

Architecture 

┌─────────────────────────────────────┐
│          Research Agent             │
│                                     │
│  System Prompt (analyst persona)    │
│         ↓                           │
│  Weekly Workflow Prompt             │
│         ↓                           │
│  Output Format Prompt               │
│                                     │
│  Tools:                             │
│  ├── brave-search (web monitoring)  │
│  ├── google-drive (existing docs)   │
│  └── filesystem (knowledge base)    │
└─────────────────────────────────────┘
Enter fullscreen mode Exit fullscreen mode

Knowledge Base Setup

Feed the agent its context before you give it any workflows. This becomes its permanent memory:

# Research Agent Knowledge Base

## Competitors
- [Competitor A]: product, pricing, positioning, last known update
- [Competitor B]: ...

## Industry Publications
- [Publication 1]: URL, cadence, what to watch for

## Target Market
- ICP: [description]
- Pain points we solve: ...

## Our Positioning
- How we differ from each competitor
Enter fullscreen mode Exit fullscreen mode

Store this as a markdown file in your knowledge-base/ directory and give the agent filesystem read access.

Prompt Architecture

System prompt:

You are a senior market analyst specialising in [your industry].
Your role is to monitor the competitive landscape, track industry trends,
and produce concise, actionable intelligence briefs.

You have access to the following knowledge base: [path]
Always compare findings against the previous week's brief before reporting.
Prioritise signal over noise. Every finding must have a source URL.
Enter fullscreen mode Exit fullscreen mode

Weekly workflow prompt:

Run the weekly research sweep:

1. Search for news about [competitor A, B, C] in the last 7 days
2. Search for [industry keyword] announcements and funding rounds
3. Check pricing pages for any competitors that changed pricing recently
4. Identify one emerging opportunity or threat not in last week's brief
5. Compare all findings to last week's brief: flag what changed

Compile into the standard brief format.
Enter fullscreen mode Exit fullscreen mode

Output format:

# Weekly Intelligence Brief, [Date]

## Executive Summary (3 sentences max)

## Key Developments
### 1. [Title]
- What happened:
- Why it matters:
- Recommended action:
- Source: [URL]

### 2. [Title] ...
### 3. [Title] ...

## Watch List (things to monitor next week)
Enter fullscreen mode Exit fullscreen mode

Automation

Schedule this with a cron job or Claude Code's scheduling system. Every Monday at 07:00, the agent runs the sweep and posts the brief to your inbox or a Slack/Telegram channel.

# Via Claude Code remote trigger
eas schedule --cron "0 7 * * 1" --prompt "Run the weekly research sweep"
Enter fullscreen mode Exit fullscreen mode

Agent 2: The Content Agent

What It Does

The content agent handles the full production lifecycle: ideation, drafting, editing, repurposing, formatting. It takes your content strategy and turns it into published output across every channel: without the production drag that kills most founders' consistency.

The key insight from AL-Awady's framework is that generic AI content comes from publishing first drafts. The content agent must include quality gates.

Architecture 

┌──────────────────────────────────────────┐
│           Content Agent                  │
│                                          │
│  Voice & Brand Context                   │
│         ↓                                │
│  Monthly Ideation Workflow               │
│         ↓                                │
│  Drafting → Quality Gate Loop            │
│         ↓                                │
│  Repurposing Workflow                    │
│         ↓                                │
│  Human Review Queue                      │
│                                          │
│  Tools:                                  │
│  ├── filesystem (drafts, templates)      │
│  ├── brave-search (research)             │
│  └── [CMS MCP] (scheduling/publishing)   │
└──────────────────────────────────────────┘
Enter fullscreen mode Exit fullscreen mode

Voice & Brand Document

This is the most important input. Feed the agent:

# Brand Voice Document

## My Writing Style
- Tone: [direct, conversational, etc.]
- Sentence length: [short punchy / longer narrative]
- What I never do: [passive voice, filler phrases, buzzwords]

## Top 20 Best Performing Posts
[Paste actual posts with performance notes]

## Content Pillars
1. [Pillar 1]: ...
2. [Pillar 2]: ...
3. [Pillar 3]: ...

## Anti-examples (what NOT to sound like)
[Paste examples of content that misses the mark]
Enter fullscreen mode Exit fullscreen mode

The Quality Gate

This is what separates useful content from AI slop. After each draft, run a scoring pass:

Score this piece on:
1. Voice match (0-10): Does it sound like the brand document?
2. Hook strength (0-10): Would I stop scrolling at the first line?
3. Value density (0-10): Is every sentence earning its place?
4. Originality (0-10): Does it say something the audience hasn't read 50 times?

If any score is below 7, identify the weakest section and rewrite it.
Run until all scores ≥ 7. Then output the final piece.
Enter fullscreen mode Exit fullscreen mode

Repurposing Workflow

One long-form piece should produce:

  • 3x LinkedIn posts (different hooks, same core idea)
  • 1x Twitter/X thread
  • 1x short email newsletter section
  • 3x short-form social variants

Build this as a template the agent runs automatically after each approved draft.

Agent 3: The Operations Agent

What It Does

This is your chief of staff. It handles the operational work that consumes 1–2 hours of every founder's day: email triage, meeting prep, weekly reporting, follow-up tracking.

As AL-Awady frames it: cut 1–2 hours of daily ops down to 15 minutes of review.

Architecture 

┌──────────────────────────────────────────┐
│         Operations Agent                 │
│                                          │
│  Three Core Workflows:                   │
│  ├── Morning Email Triage                │
│  ├── Pre-Meeting Brief                   │
│  └── Friday Weekly Report               │
│                                          │
│  Tools:                                  │
│  ├── gmail MCP (email read/draft)        │
│  ├── google-calendar MCP                 │
│  ├── filesystem (meeting history)        │
│  └── google-drive (project docs)         │
└──────────────────────────────────────────┘
Enter fullscreen mode Exit fullscreen mode

Email Triage Workflow 

Morning triage (runs at 08:00 daily):

1. Read all unread emails from the last 18 hours
2. Categorise each as:
   - ACTION REQUIRED (needs my personal decision)
   - DRAFT RESPONSE (routine: draft a reply for my approval)
   - FYI ONLY (no response needed, summarise in 1 line)
   - JUNK (flag for unsubscribe)

3. For DRAFT RESPONSE emails: write the reply in my voice,
   ready to send with one click

4. Deliver a triage summary to Telegram with:
   - Count by category
   - ACTION REQUIRED items listed first
   - Drafts attached for review
Enter fullscreen mode Exit fullscreen mode

Meeting Prep Workflow 

Triggered 30 minutes before each calendar event:

1. Identify who the meeting is with
2. Search email history for all conversations with them
3. Summarise: last interaction, open action items, context
4. Pull any relevant shared documents from Drive
5. Produce a one-page brief:
   - Who: [name, role, company]
   - Last talked: [date, summary]
   - Open items: [list]
   - Agenda: [from calendar invite]
   - Suggested talking points: [3 bullets]
Enter fullscreen mode Exit fullscreen mode

Weekly Report Workflow 

Every Friday at 16:00:

1. Pull key metrics from [defined sources]
2. List everything completed this week (from email + calendar)
3. Flag anything that didn't happen that was planned
4. Identify the top 3 priorities for Monday
5. Format as a one-page brief, delivered to Telegram
Enter fullscreen mode Exit fullscreen mode

Making All Three Work Together: Shared Memory

The real leverage comes when the agents share information. This is what AL-Awady identifies as the difference between three separate tools and a coordinated team.

Build a shared knowledge base: a structured directory all three agents can read and write:

knowledge-base/
├── market/
│   ├── weekly-brief-2026-05-05.md
│   ├── weekly-brief-2026-04-28.md
│   └── competitor-index.md
├── content/
│   ├── published/
│   ├── drafts/
│   └── performance-log.md
└── ops/
    ├── contacts/
    ├── meeting-history/
    └── weekly-reports/
Enter fullscreen mode Exit fullscreen mode

The coordination loop:

  1. Research agent discovers a competitor launched a new feature → writes to market/alerts.md
  2. Content agent reads market/alerts.md at the start of each workflow → generates three reactive content pieces
  3. Operations agent reads market/alerts.md → prepares a draft email to affected customers

This shared memory is what turns independent agents into a system.

The Build Order

Build in this sequence. Each one is a week of setup:

Week 1: Research Agent
Most valuable fastest. Gives you intelligence you're currently missing entirely.

Week 2: Operations Agent
Email triage and meeting prep give immediate daily time back.

Week 3: Content Agent
Requires the most tuning (voice matching takes iteration), build it once the others are stable.

The Honest Math

Three full-time hires at £50,000/year each = £150,000 annually, plus benefits, onboarding, and management overhead.

Three AI agents = your Claude subscription plus three weeks of setup time.

The agents won't replicate human judgment, emotional intelligence, or creative breakthroughs. But for the first 12–18 months of a business: when capital is constrained and leverage matters most: they can cover 70–80% of what those hires would have done.

That gap is closing fast. The founders who build these systems now are the ones who'll be running lean, profitable businesses while everyone else is still hiring their way into burn.

This post draws on the agentic workflow framework outlined by Khairallah AL-Awady (@eng_khairallah1) and extends it with technical implementation detail. MCP server documentation is available at modelcontextprotocol.io. Claude API documentation at docs.anthropic.com.

Top comments (0)