Building Satellites at Amazon Taught Me Traditional Planning Is Broken 

When I joined Amazon Leo (formerly Project Kuiper) to help build their satellite manufacturing supply chain, I expected to apply the same world-class planning processes I had refined at Microsoft and Starbucks. Instead, I stepped into an environment that broke my assumptions about how planning should work.  

The only viable model was one that updated as fast as the design changed – a continuous and design-linked approach – one now emerging across industries.

Lessons from Amazon Leo

Within weeks at Leo, traditional S&OP cycles and an obsession with forecast accuracy collapsed under the realities of building a next-generation satellite network. As Rajeev Badyal, VP and head of the program, put it, “We have set out to design the most advanced satellite network ever built, and we have created the whole thing in house at Amazon.” 

In a system evolving daily, a 95% accurate forecast for a design that would be obsolete by the end of the week did not reduce risk. It created it. We were locking in procurement commitments for components we would never use, creating the exact inventory exposure the planning process was supposed to prevent. 

This is not just a space problem. A Zero100 member told me recently that its newest product line behaves more like fast fashion than anything in its tech portfolio. Its planning process cannot keep up with the rate of design change, customer feedback, and rapid iteration.

And this shift from cadence-based planning to continuous planning isn’t an edge case - the gap between how businesses operate and how supply chains respond has become one of the largest execution risks in global operations. 

Why Adaptability Now Matters More Than Accuracy

In fast-moving environments, static precision loses value. Three principles from my time at Leo apply directly to modern supply chains:

• Speed outperformed precision. A workable plan delivered today created more value than a perfect one delivered a week later.
• Systems handled routine adjustments. When engineering updated a design, procurement and planning responded immediately. Shared data supported by an end-to-end tech stack replaced approval chains. Planning became a continuous capability, not a calendar event.
• Constraints anchored everything. Launch windows, test requirements, safety standards, and physics defined what could not move. Everything else flexed around these immoveable boundaries.

These patterns mirror the challenges companies face today. 

Agentic AI: The Next Layer of Continuous Planning

The continuous planning model we built at Leo through necessity is now becoming accessible to any organization through agentic AI. Zero100 research shows how these systems mirror what our teams did manually: monitor conditions, analyze signals, generate scenarios, and act within guardrails. But instead of requiring teams of engineers and planners working around the clock, agentic systems operate as autonomous teammates. offering capabilities like:

• Rebalancing inventory in real time
• Regenerating production schedules in seconds (vs daily scrambles)
• Identifying emerging constraints before they cascade
• Running parallel scenarios continuously
• Escalating only exceptions that require human judgment

These can now scale effortlessly across thousands of SKUs and suppliers, and are how continuous planning moves from edge case to industry standard.

Fusion Teams: The Talent Model for Modern Planning

The team at Leo didn't look like a traditional planning organization. Some of our strongest problem solvers came from engineering or software backgrounds. Traditional planning expertise remained vital, but it had to blend with systems thinking and comfort with rapid experimentation. 

At Zero100, we call these fusion teams: collaborative groups that blend engineering, data science, product management, and operational expertise. They translate business needs into technical requirements, own outcomes end-to-end, and work in continuous loops with autonomous systems to drive measurable business results. 

Our analysis of supply chain planning job descriptions reveals a striking gap. While leading companies recruit for AI orchestration and autonomous planning capabilities, 44% of planning roles still prioritize Excel-based S&OP skills. But organizations that shift their hiring from Excel Jugglers to Orchestrators will own the future of planning.  

Building a Continuous Planning Model

The lessons for planning from my time in satellite manufacturing?

1. Speed over Accuracy 
Measure planning by how fast it updates, not by how precise it was on day one. A good plan today beats a perfect plan next week. 

2. Constraints as Anchors 
Define what can’t move—launch windows, regulatory deadlines, physics—and build flexibility everywhere else. Constraints clarify; everything else is negotiable. 

3. Agents in the Loop, Humans for Strategy 
Use agentic AI for routine adjustments and parallel scenarios. Let humans focus on strategy, exceptions, and the decisions that actually need judgment. 

All of this is the ongoing governance work of fusion teams. 

The companies that thrive will not be the ones with the most sophisticated traditional planning process. They will be the ones that rebuild planning to match the speed of the world in which they operate.