From planning systems to self-optimizing enterprises

Author

Thomas Steirer is Chief Technology Officer (CTO) at Nagarro. His focus is on developing scalable and sustainable solutions that are primarily designed to deliver valuable information.

Why the smartest engineering teams invest as much in the factory as the product

Here’s something that keeps coming up in conversations with engineering leaders, and it’s worth diving into: The best technology teams in the world now spend roughly half their time building the system that builds the product, rather than the product itself.

That ratio feels wrong at first. Fifty percent of your engineering capacity not focused on features or revenue-generating output? It sounds like overhead. But there are tangible assets and practices for the discipline behind this allocation, for example: Compound Engineering. Once you understand the logic, it becomes hard to argue against. It is, of course, only one manifestation out of many AI-first practices and frameworks across many disciplines and domains, but they all follow similar principles, focused on what we call "meta-engineering" or "second-order optimization." We'll use Compound Engineering as one representative example of this broader class of concepts.

Roughly 50% of engineering effort should go toward improving the systems, tooling, and platforms that produce the product, not the product features themselves. This is the general meta-engineering principle: every investment in the meta-system compounds over time. Better CI/CD pipelines, smarter testing frameworks, and more capable internal platforms may not show up in a release note, but they accelerate everything that does. Compound Engineering is one framework that operationalizes this principle, particularly within software development teams.

The opposite of this meta-engineering approach is the Feature Factory: teams shipping features at maximum velocity while their underlying systems quietly atrophy. Output looks impressive quarter by quarter, but capability stagnates. The factory rusts while the shelves stay stocked. It’s a pattern most engineering leaders will recognize, and it’s exactly what the organizations pulling ahead have chosen to reject.

Once you look at what’s actually happening inside organizations like Tesla, JPMorgan Chase, and Siemens, the shift becomes concrete. This is the move from planning systems to self-optimizing enterprises, and it changes what it means to lead a technology organization.

One of the biggest shifts in modern business thinking is realizing that competitive advantage often comes less from the product itself and more from the system that produces it. The organizations pulling ahead aren’t just investing in what they sell, they’re investing in the engines that allow them to create, deliver, and improve value at scale.

The pattern

Different industries. Different products. Same principle.

Amazon built a fulfillment engine. Zara built a sensing-and-production engine. BYD built an integrated industrial engine.

The organizations creating durable advantage aren’t just improving outputs. They’re improving the systems that generate those outputs. Every enhancement to the warehouse, supply chain, manufacturing platform, or operating model compounds across everything that follows.

This is meta-engineering in practice. Compound Engineering is one framework that codifies these principles specifically for software development teams. The most forward-looking organizations treat their production systems as strategic assets under continuous development, not as fixed infrastructure to be maintained, but as living systems to be measured, optimized, and continuously improved.

When banks start thinking like manufacturers

If this were only a manufacturing insight, we could file it under “interesting, but not my problem.” It’s not. The same pattern is reshaping financial services.

JPMorgan Chase commits $18 billion annually to technology, with roughly $2 billion directed specifically toward AI and machine learning. But what’s genuinely interesting isn’t the spend. It’s where that spend goes. A significant share funds platforms like OmniAI, a centralized ML “factory floor” that standardizes the entire machine learning lifecycle, from research to production, across more than 450 AI use cases. There’s also LLM Suite, a model-agnostic generative AI platform now serving more than 200,000 employees and recognized as American Banker’s Innovation of the Year in 2025.

What makes this meta-engineering story, not just a technology investment story, is that JPMorgan’s engineers don’t build trading algorithms and fraud models at all. They build the platform that produces trading algorithms and fraud models. They build the tooling that governs, monitors, and scales those models. They build PRBuddy, an internal tool that uses AI to automate code reviews. Each platform improvement multiplies across hundreds of use cases simultaneously. The effort goes into the factory, not the widgets coming off the line.

From planning to self-optimization

Meta-engineering represents the broader shift toward self-improving systems and second-order optimization. Compound Engineering is one architectural approach that can help organizations realize this vision.

Now take this one step further. What happens when the factory doesn’t just produce efficiently, but improves itself, sensing, learning, and adjusting without waiting for human direction?

Siemens has been answering that question at its Amberg and Erlangen plants. Their approach centers on the comprehensive digital twin: a virtual replica of the entire production lifecycle that doesn’t merely simulate but actively optimizes operations in real time. Their Executable Digital Twins run on industrial PCs using live sensor data, enabling physics-based simulation and autonomous decision-making directly on the factory floor, with no cloud round-trip required.

The outcomes at Amberg are impressive: 99.9% production quality, a 20% increase in productivity, a 30% reduction in unplanned downtime, and more than $35 million in annual operational savings. Their roadmap explicitly targets “self-healing systems,” enabling automated defect correction without human intervention.

What makes Siemens instructive isn’t just the efficiency. It’s that these systems are genuinely agentic. Traditional manufacturing relies on human planners to design production schedules, quality checks, and maintenance windows. Siemens has replaced much of that with systems that continuously sense, simulate, and adjust on their own. The factory optimizes itself.

When your digital twin is making real-time production decisions faster and more accurately than your best process engineer, you have crossed into something qualitatively different. You have a self-optimizing enterprise: an agentic system where the factory doesn’t just execute plans, but generates, tests, and refines them autonomously.

This is the next stage of meta-engineering. Compound Engineering and similar forward-facing AI-first approaches are frameworks that help organizations operationalize this shift. The meta-system, at its core, doesn’t focus on accelerating day-to-day work. It focuses on learning, and on persisting the gained learnings to improve itself. The primary investment no longer goes into running and maintaining the factory. It goes into continuously improving how the factory itself thinks.

The pattern across sectors

These examples aren’t outliers. Goldman Sachs is piloting autonomous AI coding agents alongside 12,000 developers, projecting 3–4x productivity gains. Walmart has achieved 98% automation in select distribution centers. According to Bain & Company research, one consumer products firm cut its forecasting cycle from two weeks to two hours while reaching 97% accuracy.  Leading organizations are moving beyond automating individual tasks to building agentic systems that continuously optimize the infrastructure of work itself.

Making the shift practical

The challenge, of course, is that most enterprises aren’t Tesla or JPMorgan. They don’t have $18 billion technology budgets. The strategic question for most leadership teams is simple: how do you adopt this mindset at an achievable scale?

This is where frameworks that focus on where intelligence gets stuck become genuinely useful. Nagarro’s Fluidic Intelligence approach, for instance, operationalizes these meta-engineering principles. Like Compound Engineering, it provides a structured way to invest in the system that builds the product. It begins by diagnosing friction across three layers:

From there, advisory designs the factory blueprint. Forge builds and integrates the machinery using platform-agnostic accelerators. Augmented engineering teams embed agentic collaboration into the development lifecycle, operating and continuously improving the system.

What’s notable isn’t any single technology choice. It’s the architecture of the transformation. The approach treats the system that produces intelligence with the same rigor as the intelligence being produced. Invest in the meta-system, and every downstream output improves.

Platform-agnostic design matters here, too. Lock-in is the enemy of compounding. If every improvement to your AI capability is trapped within a single vendor’s ecosystem, you lose the portability that allows those improvements to multiply across contexts.