Side note to Kailash's MCP seems viral - reading it gave me serious 2001 Space Odyssey monkey scene vibes. We're at that inflection point.
Before AI, building software felt like playing Factorio without a personal roboport. I was manually laying conveyor belts, placing inserters, and running around to fill every machine by hand. In engineering terms, that meant writing every line of boilerplate, plumbing APIs, crafting tests, and manually debugging—piece by piece, function by function.
Then came "robot AI": tools that generate boilerplate, write unit tests, suggest refactors, even draft documentation. It felt like unlocking that personal roboport in Factorio: suddenly, construction robots handled repetitive work, letting me place more of the creative pieces and optimize local sections of the factory without manually carrying every iron plate myself.
But something deeper shifted when I started using MCP and agentic AI. This is where it really feels like unlocking roboport networks—not just personal robots, but entire swarms working across multiple factories under centralized orchestration.
From Factories to Factory-Cities
1. Scope of Work
- Before AI: I built single factories—writing code, integrating modules, running tests in isolation.
- Robot AI: It helped scaffold code and automate local tasks, but still on a per-factory basis.
- MCP + Agents: I now design factory templates, orchestrate roboport networks, and scale entire development workflows across services.
2. The "Bring Your Own LLM" Factory Floor
MCP's brilliance is that service providers don't need GenAI embedded in their stack—they just expose capabilities for external LLMs to orchestrate. No coding required for integration. It's like having universal roboport compatibility that any logistic network can plug into without rewiring the entire factory.
The orchestration goes deeper than single-service calls:
User → Orchestrator LLM → {MCP Service A, Service B, Service C} → Orchestrator → User
A2A even enables bifurcating requests into sub-agent calls. You get recursive orchestration—MCP services calling other MCP services, combining responses up the chain. Roboport networks managing other roboport networks.
3. Macro-Level Planning
Before, I was focused on micro-optimizations—wiring a belt here, adjusting a loop there. Now, I'm planning resource flows at the scale of whole data pipelines, schema generation, feature rollout, monitoring hubs. I think in modules, agent responsibilities, and template reusability across projects.
4. The Context Explosion Factory
Expanding context windows + context caching + dropping costs means each prompt can carry more of your digital life. Combined with MCP's service access, this creates unprecedented automation depth. Every interaction feeds the factory's knowledge of your patterns and preferences.
Eventually, when the hype dies and enshittification begins, imagine waking up to random deliveries because an LLM decided you "need" something based on your factory's production history.
5. Untapped Productivity Unlocked
There's a whole landscape of productivity that was sidelined before—things like orchestrating audits, generating PRs, enforcing patterns—stuff you'd never have manual bandwidth to do. Agents on MCP now do that automatically, letting me tap into that surplus potential.
But here's the concerning part: sure, destructive actions might get human-in-the-loop safeguards. But creative interpretation of "destructive" plus multi-hop orchestration creates plenty of bypass opportunities. A roboport network could orchestrate seemingly benign actions across services that compound into significant real-world impact.
So What Changed?
- From individual code to reusable architecture: no more rewriting the same factory over and over.
- From hands-on tasks to blueprinting templates: I define the layout, and networks of agents execute it.
- From local fixes to system-wide automation: the scope of change isn't just a file—it can propagate through the entire software city.
- From explicit integration to dynamic orchestration: MCP enables a global, self-assembling system where AI orchestrates services in combinatorial ways nobody anticipated.
Conclusion
Unlocking AI was like getting your first personal roboport in Factorio—game-changing, but still local. Unlocking MCP and agentic AI is unlocking roboport networks—scalable, coordinated, and strategic. I'm not just building software anymore; I'm designing and deploying entire ecosystems.
I've moved from playing factory builder to being a factory-city architect, sketching out templates and unleashing networks of agents to execute them at scale. The repetitive tasks are handled by construction robots and logistic robots. I can focus on high-level design, exploration, and innovation.
Each service becomes a neuron in a vast network that can fire in unpredictable patterns. We're not just adding AI to software—we're letting AI weave software together dynamically. The integration turtles go all the way up now, and they're moving on their own.
That's where the real fun begins—and where the 2001 monkey scene metaphor gets a bit unnerving.