Search for a command to run...
No comments yet. Be the first to comment.
Privacy isn't dead. It's just being rebuilt. The old internet promised us "move fast and break things." We moved fast. We broke things. Now we're realizing what we broke: privacy. Every LLM you use wa
Coding is changing faster than you think. Not just "AI helps with autocomplete." AI is becoming the primary developer. You become the reviewer. The Skill Inversion Old model (2020): Human writes code.
You've learned basic RAG. You embed documents. You retrieve them. You feed them to an LLM. It generates an answer. Then you deploy it. And it breaks. Because the real world isn't simple retrieval. Adv
The biggest AI models aren't winning anymore. GPT-4: 1.8 trillion parameters Claude 3 Opus: 200+ billion parameters Claude 3.5 Sonnet: Unknown but smaller Llama 3.1: 405 billion parameters PhiMoe (ope
Muhammad Zulqarnain | Full Stack AI Engineer & Geospatial Developer
16 posts
A blog by Muhammad Zulqarnain — Full Stack AI Engineer & Geospatial Developer based in Turku, Finland. I write about RAG systems, LLMs, Prompt Engineering, Next.js, TypeScript, and geospatial development. Practical insights, deep dives, and real-world AI solutions.
Single AI agents are limited. They make mistakes. They miss context. They take wrong paths.
Multi-agent systems don't have these problems.
They have different problems.
But better ones.
One AI doing everything:
Research
Analysis
Writing
Code generation
Fact-checking
Result: Mediocre at everything.
Specialized agents:
Research agent gets documents
Analysis agent finds patterns
Writer agent composes
Coder agent generates
Verifier agent checks
Result: Excellent output.
The Orchestrator
Main agent decides:
Which agents to use
In what order
What data to pass
When to stop
The Specialists
Each agent:
Has specific training
Has access to specific tools
Has clear responsibilities
Validates its own output
Agent 1: Extractor Pulls structured data from documents
Agent 2: Classifier Categorizes the data
Agent 3: Fact-Checker Verifies claims against knowledge base
Agent 4: Summarizer Creates executive summary
Agent 5: Quality Control Validates everything
Each agent does one thing well.
Orchestrator coordinates.
Output: Reliable, structured, verified.
Challenge 1: Deadlocks
Agent A waiting for Agent B Agent B waiting for Agent C Agent C waiting for Agent A
System hangs.
Solution: Timeout + fallback
Challenge 2: Conflicting Decisions
Agent says "hire this person" Agent 2 says "don't trust this person"
Who's right?
Solution: Explicit voting or hierarchy
Challenge 3: Latency
5 agents = 5x the API calls (maybe)
Could be 5x slower.
Solution: Parallel execution, caching
Code Generation + Testing
Agent 1: Generates code
Agent 2: Writes tests
Agent 3: Runs tests
If fails: Agent 1 fixes
Loop until passing
Result: Code that actually works
Customer Support Triage
Agent 1: Classifies issue
Agent 2: Retrieves relevant docs
Agent 3: Drafts response
Agent 4: Routes to human if needed
Result: Faster resolution
Content Creation
Agent 1: Research
Agent 2: Outline
Agent 3: Draft
Agent 4: Edit
Agent 5: Format
Result: Publication-ready content
Hierarchical Agents
CEO agent delegates to:
Research agent
Analysis agent
Writing agent
Each reports back. CEO synthesizes.
Debate Agents
Agent A argues for solution X Agent B argues for solution Y Judge agent decides
Better reasoning than single agent.
Specialist Ensembles
3 coding agents generate Best 2 outputs selected Combined
Quality higher than single agent.
Step 1: Define Agents
What specific things do you need done? Create an agent for each.
Step 2: Define Handoffs
Agent A completes Agent B starts What data passes between?
Step 3: Add Validation
Each agent checks its output Passes confidence score Orchestrator uses score
Step 4: Error Handling
Agent fails? Retry with different approach Escalate to human
Step 5: Monitor
Which agents are slow? Which make mistakes? Optimize
LangGraph
Define agent workflows as graphs Visualize Debug
AutoGen (Microsoft)
Multiple agents converse Reach consensus Execute plan
Crew AI
Roles. Tools. Tasks. Coordinated execution
Linear scaling
1 agent: 100ms 2 agents: 150ms (parallel) 5 agents: 250ms
Not 500ms because parallelization
Exponential complexity
2 agents: 1 handoff 3 agents: 6 possible paths 4 agents: 24 paths 5 agents: 120 paths
Optimization becomes hard.
Use:
Complex workflows
Need specialization
Want reliability
Different tools needed
Don't use:
Simple tasks
Single agent sufficient
Latency critical (and parallelization impossible)
1. Agent Marketplaces
"Buy an agent" Specialized agents you plug in Communicate via standard interfaces
2. Self-Improving Agents
Agents learn from feedback Improve over time Auto-optimize handoffs
3. Hierarchical AI Organizations
Not just agent teams Agents managing agents Organizational structure
4. Cross-Company Agents
Your agent talks to their agent Auto-negotiate Auto-execute
Multi-agent is harder than single agent.
But output is better.
Coordination is the challenge.
Solve coordination, you win.s
