Experimentation Is Becoming a Core Principle of Engineering

For years, engineering leaders treated the 70/20/10 rule as a kind of operational law: a clean, rational allocation of effort that balanced stability with innovation. In theory, it worked. In practice, it rarely held.

The 70% dedicated to core systems (maintenance, bug fixes, incremental updates) had a tendency to expand. Quietly, then completely. What was meant to anchor the business became the business. Engineering teams found themselves trapped in a loop of keeping the lights on, with experimentation relegated to side projects or the occasional “innovation sprint” that never quite made it into production.

That dynamic is now changing. And it’s not because companies suddenly value innovation more. It’s because AI is fundamentally reshaping the structure of building, maintaining, and experimenting with technology.

We’re not just revisiting the 70/20/10 model. We’re watching it evolve in real-time.

The Framework That Defined Engineering Discipline

The original appeal of the 70/20/10 model, popularized by Google, was its clarity.

Seventy percent of resources went to the core business: the systems and features that generated revenue and served existing customers. Twenty percent supported adjacent projects: extensions of what already worked, applied to new markets or use cases. The remaining ten percent was reserved for transformational ideas: high-risk, high-reward bets that could redefine the business.

Organizations that executed this well tended to outperform. They didn’t only chase the next big thing—they protected their foundation while leaving room for discovery.

But this model assumed something that no longer holds true: that engineering effort is linear and largely human-bound.

From Execution to Judgment: Why the Model is Shifting

What’s changed isn’t just velocity; it’s the nature of engineering leverage.

In the traditional 70/20/10 model, the high “cost of discovery” meant that experimentation was often a late-stage luxury. Because execution was the primary bottleneck, you had to front-load your certainty before committing resources. Today, AI-driven leverage has inverted that logic. Experimentation is no longer what we do after we’ve made a decision—it’s the tool we use to make the decision.

By lowering the overhead of building and testing, we can now use experimentation as a mechanism for prioritization. This shift manifests in three key ways:

• Validation at the Speed of Thought: Because we can build faster, we must use experimentation to pivot faster. We experiment more not to increase our output, but to decrease the time it takes to find the high-impact solutions that align with our core business.
• High-Fidelity Testing: Because AI has made it significantly faster and easier to develop, we can move away from isolated prototypes or “ahead of time” validation. Experimentation can now be more complete, using live features to test ideas against real-world complexity.
• The Judgment Loop: Every experiment provides data that sharpens our strategic priorities. More experimentation doesn’t lead to a fragmented product vision; rather, it provides the empirical evidence needed to double down on the right 70%.

The result is an accelerating cycle: higher capacity leads to more experiments, which provides more data, which leads to better judgment. We aren’t experimenting more because we have extra time—we are doing it because, in an AI-driven world, experimentation is a way to ensure our increased capacity is being pointed at the right problems.

The New Operating System for Engineering Teams

If AI has accelerated the pace of development, it has also compressed expectations. Product cycles are shorter. Releases are more frequent. Feedback loops are tighter. But the more profound shift is structural. Engineering teams are no longer just building software. They are orchestrating systems of intelligence.

The traditional workflow (pull a ticket, write code, test, debug) is giving way to something closer to supervision and direction. Engineers are increasingly responsible for deciding which problems to solve, which agents should solve them, and how to evaluate the results. In that sense, every engineer is becoming a manager. Not only of people, but of agents.

A single engineer can now coordinate dozens—or hundreds—of parallel problem-solving processes. One agent writes code. Another tests it. A third evaluates performance. The engineer’s role is to guide, review, and refine. It’s less about execution and more about judgment. This shift has cascading effects across the organization:

• Silos break down. When workflows are mediated by AI, rigid team boundaries become less relevant. Product, engineering, and data functions begin to merge into fluid, outcome-driven units.
• R&D scales non-linearly. Instead of adding headcount to explore new ideas, teams deploy fleets of agents. Experimentation scales without the traditional cost curve.
• Leadership is redefined. The value of managers and senior engineers shifts away from coordination toward prioritization and technical discernment. The question is no longer “Can we build this?” but “Should we?”

In this model, the bottleneck is no longer execution. It’s decision-making.

Building a Culture That Can Keep Up

Technology alone doesn’t create a culture of experimentation. In many cases, it exposes the absence of one. Organizations that succeed in this new environment are doing a few things differently.

They are dismantling legacy structures that assume work must be sequential and human-bound. They are investing in AI-native workflows, where agents are embedded into every stage of development—from ideation to deployment. And they are also redefining what it means to be an engineer.

The modern engineer is part architect, part reviewer, part strategist. Writing code is no longer the primary source of value. Reviewing machine-generated output, ensuring correctness, designing for scale, and interpreting context—these are becoming the core competencies.

Just as importantly, leaders are aligning their teams around a shared understanding of how autonomous systems should be used. Experimentation without direction leads to noise. Experimentation guided by clear prioritization prevails.

The Return to Human-Centric Engineering

There’s an irony in all of this. As engineering becomes more automated, it’s also becoming more human. When AI takes over the repetitive and the mechanical, it creates space for something else: curiosity.

Teams spend less time digging through legacy systems and more time building. A proof of concept that once took a week may now be assembled in an afternoon. Not perfectly, but well enough to test, iterate, and learn. 

That speed changes behavior. It lowers the cost of being wrong. It encourages teams to explore unconventional ideas. It makes experimentation not only possible, but practical.

And that’s where leaner, more agile organizations gain an advantage. They don’t win by outspending larger competitors. They win by out-experimenting them, by being fast enough to test ideas that others don’t have the agility to pursue.

The 70/20/10 rule was designed for a world where resources were constrained, and predictability mattered most. We’re entering a different world—one where adaptability, speed, and creativity define success. In that world, the companies that thrive won’t be the ones that protect their 70%. They’ll be the ones who rethink it entirely.