DEV Community: Jovan Marinovic

AI Writes Code, But Who Coordinates the AI? The Missing Accountability Layer

Jovan Marinovic — Thu, 30 Jul 2026 09:47:39 +0000

As AI agents write more code and make more decisions, the accountability question isn't just philosophical — it's an engineering problem.

The Article That Sparked This

I recently read @lizziepika's excellent article "Agents write code, but they don't remember" and it resonated deeply with challenges I've been solving in production.

The responsibility question raised here extends directly into multi-agent systems. When multiple AI agents produce a result, which one is accountable? You need audit trails.

The Core Problem: State Coordination

Here's what most multi-agent discussions miss: the frameworks are great at individual agent capabilities. LangChain gives you chains, AutoGen gives you conversations, CrewAI gives you roles. But when these agents need to share state — that's where things silently break.

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

Accountability Requires Infrastructure

You can't assign responsibility without visibility. Full audit trails, permission gating, and atomic state tracking turn "who did this?" from a mystery into a lookup.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

How are you handling accountability in your AI agent systems? Drop your approach below!

What Every AI Agent Builder Needs to Know About State Coordination

Jovan Marinovic — Tue, 28 Jul 2026 09:52:43 +0000

After months of building multi-agent AI systems, the biggest lesson: the framework doesn't matter as much as the coordination layer.

The Article That Sparked This

I recently read @sylwia-lask's excellent article "The Dirty Secret Behind AI Agents (Demo 🚀)" and it resonated deeply with challenges I've been solving in production.

This article touches on a challenge we've been obsessing over: how to make AI agents work together reliably without custom glue code for every interaction.

The Core Problem: State Coordination

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

The Hard Part Isn't the Model

Better models won't fix coordination problems. You need purpose-built infrastructure for state management, conflict resolution, and cross-agent communication.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

Building multi-agent systems? I'd love to hear about your architecture — let's compare notes in the comments!

MCP Is a Great Start — But Multi-Agent Production Needs More

Jovan Marinovic — Thu, 23 Jul 2026 09:45:15 +0000

The Model Context Protocol has transformed how we connect AI to tools. But connecting agents to tools is only half the battle — connecting agents to each other is where the real challenge begins.

The Article That Sparked This

I recently read @mattlewandowski93's excellent article "I gave Claude six months of our retros. It found three things I'd missed." and it resonated deeply with challenges I've been solving in production.

This post highlights exactly what makes MCP powerful. Where I want to extend the conversation is: what happens when you have 3, 5, or 10 MCP-powered agents all sharing context?

The Core Problem: State Coordination

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

MCP + Network-AI: The Full Stack

MCP handles the agent-to-tool connection brilliantly. Network-AI adds the agent-to-agent coordination layer. Together, they give you a full production stack for multi-agent systems.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

Running MCP agents in production? I'd love to hear what coordination challenges you've hit — drop a comment!

What Every AI Agent Builder Needs to Know About State Coordination

Jovan Marinovic — Tue, 21 Jul 2026 09:49:41 +0000

After months of building multi-agent AI systems, the biggest lesson: the framework doesn't matter as much as the coordination layer.

The Article That Sparked This

I recently read @debs_obrien's excellent article "How I Documented an Entire Product in 4 Days with an AI Agent" and it resonated deeply with challenges I've been solving in production.

This article touches on a challenge we've been obsessing over: how to make AI agents work together reliably without custom glue code for every interaction.

The Core Problem: State Coordination

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

The Hard Part Isn't the Model

Better models won't fix coordination problems. You need purpose-built infrastructure for state management, conflict resolution, and cross-agent communication.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

Building multi-agent systems? I'd love to hear about your architecture — let's compare notes in the comments!

MCP Is a Great Start — But Multi-Agent Production Needs More

Jovan Marinovic — Thu, 16 Jul 2026 09:37:59 +0000

The Model Context Protocol has transformed how we connect AI to tools. But connecting agents to tools is only half the battle — connecting agents to each other is where the real challenge begins.

The Article That Sparked This

I recently read @dannwaneri's excellent article "MCP Just Landed on Your Phone: What Google AI Edge Gallery Actually Does" and it resonated deeply with challenges I've been solving in production.

This post highlights exactly what makes MCP powerful. Where I want to extend the conversation is: what happens when you have 3, 5, or 10 MCP-powered agents all sharing context?

The Core Problem: State Coordination

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

MCP + Network-AI: The Full Stack

MCP handles the agent-to-tool connection brilliantly. Network-AI adds the agent-to-agent coordination layer. Together, they give you a full production stack for multi-agent systems.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

Running MCP agents in production? I'd love to hear what coordination challenges you've hit — drop a comment!

What Every AI Agent Builder Needs to Know About State Coordination

Jovan Marinovic — Tue, 14 Jul 2026 09:34:26 +0000

After months of building multi-agent AI systems, the biggest lesson: the framework doesn't matter as much as the coordination layer.

The Article That Sparked This

I recently read @erikch's excellent article "Build Your Own AI Butler - A Scheduled Agent That Runs Itself!" and it resonated deeply with challenges I've been solving in production.

This article touches on a challenge we've been obsessing over: how to make AI agents work together reliably without custom glue code for every interaction.

The Core Problem: State Coordination

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

The Hard Part Isn't the Model

Better models won't fix coordination problems. You need purpose-built infrastructure for state management, conflict resolution, and cross-agent communication.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

Building multi-agent systems? I'd love to hear about your architecture — let's compare notes in the comments!

Why Your Multi-Agent AI System Needs Governance (Not Just Orchestration)

Jovan Marinovic — Thu, 09 Jul 2026 10:11:53 +0000

Most multi-agent frameworks focus on capabilities. Few address governance: who can do what, how conflicts are resolved, and how you maintain audit trails.

The Article That Sparked This

I recently read @ben's excellent article "From Harness Engineering to Evals: What’s Trending at AI Engineer" and it resonated deeply with challenges I've been solving in production.

The governance angle here is spot-on and massively underappreciated. In production, knowing what each agent CAN do matters as much as what it SHOULD do.

The Core Problem: State Coordination

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

Governance Is Infrastructure

Just like databases need transactions and web apps need authentication, multi-agent systems need governance. It's not optional — it's infrastructure.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

How are you handling governance in your multi-agent systems? Share your approach!

What Every AI Agent Builder Needs to Know About State Coordination

Jovan Marinovic — Tue, 07 Jul 2026 10:10:29 +0000

After months of building multi-agent AI systems, the biggest lesson: the framework doesn't matter as much as the coordination layer.

The Article That Sparked This

I recently read @ishaansehgal's excellent article "The Log Is the Agent" and it resonated deeply with challenges I've been solving in production.

This article touches on a challenge we've been obsessing over: how to make AI agents work together reliably without custom glue code for every interaction.

The Core Problem: State Coordination

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

The Hard Part Isn't the Model

Better models won't fix coordination problems. You need purpose-built infrastructure for state management, conflict resolution, and cross-agent communication.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

Building multi-agent systems? I'd love to hear about your architecture — let's compare notes in the comments!

MCP Is a Great Start — But Multi-Agent Production Needs More

Jovan Marinovic — Thu, 02 Jul 2026 09:59:51 +0000

The Model Context Protocol has transformed how we connect AI to tools. But connecting agents to tools is only half the battle — connecting agents to each other is where the real challenge begins.

The Article That Sparked This

I recently read @marcosomma's excellent article "The Model Does Not Need Memory. The Situation Does." and it resonated deeply with challenges I've been solving in production.

This post highlights exactly what makes MCP powerful. Where I want to extend the conversation is: what happens when you have 3, 5, or 10 MCP-powered agents all sharing context?

The Core Problem: State Coordination

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

MCP + Network-AI: The Full Stack

MCP handles the agent-to-tool connection brilliantly. Network-AI adds the agent-to-agent coordination layer. Together, they give you a full production stack for multi-agent systems.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

Running MCP agents in production? I'd love to hear what coordination challenges you've hit — drop a comment!

What Every AI Agent Builder Needs to Know About State Coordination

Jovan Marinovic — Tue, 30 Jun 2026 10:13:13 +0000

After months of building multi-agent AI systems, the biggest lesson: the framework doesn't matter as much as the coordination layer.

The Article That Sparked This

I recently read @aditya_007's excellent article "Hermes Mentor — A Local AI Agent That Gets You Out of Tutorial Hell" and it resonated deeply with challenges I've been solving in production.

This article touches on a challenge we've been obsessing over: how to make AI agents work together reliably without custom glue code for every interaction.

The Core Problem: State Coordination

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

The Hard Part Isn't the Model

Better models won't fix coordination problems. You need purpose-built infrastructure for state management, conflict resolution, and cross-agent communication.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

Building multi-agent systems? I'd love to hear about your architecture — let's compare notes in the comments!

Why Your Multi-Agent AI System Needs Governance (Not Just Orchestration)

Jovan Marinovic — Thu, 25 Jun 2026 10:01:36 +0000

Most multi-agent frameworks focus on capabilities. Few address governance: who can do what, how conflicts are resolved, and how you maintain audit trails.

The Article That Sparked This

I recently read @kenielzep97's excellent article "Every Step Was Allowed. The Sequence Was the Attack. (AI Memory Judgment, CLAIM-30)" and it resonated deeply with challenges I've been solving in production.

The governance angle here is spot-on and massively underappreciated. In production, knowing what each agent CAN do matters as much as what it SHOULD do.

The Core Problem: State Coordination

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

Governance Is Infrastructure

Just like databases need transactions and web apps need authentication, multi-agent systems need governance. It's not optional — it's infrastructure.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

How are you handling governance in your multi-agent systems? Share your approach!

MCP Is a Great Start — But Multi-Agent Production Needs More

Jovan Marinovic — Tue, 23 Jun 2026 10:15:41 +0000

The Model Context Protocol has transformed how we connect AI to tools. But connecting agents to tools is only half the battle — connecting agents to each other is where the real challenge begins.

The Article That Sparked This

I recently read @sbis04's excellent article "Ghost Maintainer — An AI Junior Partner for Open Source" and it resonated deeply with challenges I've been solving in production.

This post highlights exactly what makes MCP powerful. Where I want to extend the conversation is: what happens when you have 3, 5, or 10 MCP-powered agents all sharing context?

The Core Problem: State Coordination

Timeline of a Production Bug:
0ms:  Agent A reads shared context (version: 1)
5ms:  Agent B reads shared context (version: 1)  
10ms: Agent A writes new context (version: 2)
15ms: Agent B writes context (based on v1) → OVERWRITES Agent A
Result: Agent A's work is silently lost. No error thrown.

This isn't hypothetical — it's the #1 failure mode in multi-agent production systems.

How We Solved It: Network-AI

After hitting this wall repeatedly, I built Network-AI — an open-source coordination layer that sits between your agents and shared state:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│  LangChain  │  │   AutoGen   │  │   CrewAI    │
└──────┬──────┘  └──────┬──────┘  └──────┬──────┘
       │                │                │
       └────────────────┼────────────────┘
                        │
                 ┌──────▼──────┐
                 │  Network-AI │
                 │ Coordination│
                 └──────┬──────┘
                        │
                 ┌──────▼──────┐
                 │ Shared State│
                 └─────────────┘

Every state mutation goes through a propose → validate → commit cycle:

// Instead of direct writes that cause conflicts:
sharedState.set("context", agentResult); // DANGEROUS

// Network-AI makes it atomic:
await networkAI.propose("context", agentResult);
// Validates against concurrent proposals
// Resolves conflicts automatically
// Commits atomically

Key Features

🔐 Atomic State Updates — No partial writes, no silent overwrites
🤝 14 Framework Support — LangChain, AutoGen, CrewAI, MCP, A2A, OpenAI Swarm, and more
💰 Token Budget Control — Set limits per agent, prevent runaway costs
🚦 Permission Gating — Role-based access across agents
📊 Full Audit Trail — See exactly what each agent did and when

MCP + Network-AI: The Full Stack

MCP handles the agent-to-tool connection brilliantly. Network-AI adds the agent-to-agent coordination layer. Together, they give you a full production stack for multi-agent systems.

Try It

Network-AI is open source (MIT license):

👉 https://github.com/Jovancoding/Network-AI

Join our Discord community: https://discord.gg/Cab5vAxc86

Running MCP agents in production? I'd love to hear what coordination challenges you've hit — drop a comment!