Autonomous vehicles: when 90% done means nowhere near ready

1. The Promise: reimagining mobility

For decades, the idea of autonomous vehicles has captured the imagination of engineers, policymakers, and society at large. At its core, the vision is compelling: a world where mobility is seamless, efficient, and largely invisible.

In such a future, commuting time is reclaimed. Cars become extensions of our living or working spaces rather than tools requiring constant attention. Ownership models shift dramatically: vehicles no longer sit idle 95% of the time but operate continuously, transporting passengers, then repositioning themselves autonomously.

Urban landscapes evolve: fewer parking spaces, less congestion, and a more efficient use of public space.
The promise extends beyond convenience. Autonomous systems hold the potential to drastically reduce human error (the leading cause of road accidents!) thereby improving safety outcomes at scale.
It is, in every sense, a transformative vision. And for a time, it felt imminent.

2. The Reality: understanding levels of autonomy

To understand where we stand, it is essential to ground the discussion in the standardized classification of driving automation, ranging from Level 0 (no automation) to Level 5 (full autonomy).

• L0–L1: Minimal assistance (e.g., cruise control, lane keeping).
• L2: Partial automation: systems can control steering and acceleration simultaneously, but human supervision is mandatory.
• L3: Conditional automation: vehicles can handle most driving tasks under specific conditions, but human takeover is required when prompted.
• L4: High automation: vehicles can operate autonomously in defined environments without human intervention.
• L5: Full autonomy: no human input required under any conditions.

For many of us, the concept of fully autonomous driving once felt like pure science fiction. I still recall watching Total Recall and seeing robot-driven taxis navigating the city, an idea that seemed completely far-fetched at the time. Interestingly, those fictional vehicles would likely be classified today as Level 4, or possibly even Level 5, depending on whether they were constrained to specific environments or capable of operating universally.
In reality, most commercially available systems, including those from Tesla, operate at Level 2, despite frequent public perception suggesting otherwise.

Meanwhile, companies like Waymo have achieved Level 4 capabilities in tightly controlled environments. However, the leap to Level 5 remains elusive. Notably, for over a decade, industry leaders have suggested that full autonomy was “just around the corner.” Yet, year after year, that milestone has remained out of reach.

3. The 90–10 Problem: when progress stalls at the finish line

This gap between expectation and reality can be understood through the lens of the “90–10 problem,” a well-known engineering principle often cited in discussions of complex systems.

The principle is deceptively simple: The first 90% of a project often takes 10% of the time, while the final 10% consumes the remaining 90%.

In the context of autonomous vehicles, the industry has largely solved the “easy” part: structured environments, predictable conditions, controlled variables. This is the world of highways, clear weather, and well-marked roads.
The remaining 10% is where complexity explodes:

• Unpredictable human behavior: a pedestrian suddenly crossing outside a designated crosswalk while looking at their phone
• Edge cases (rare but critical scenarios): a mattress falling off a truck on a highway or an unexpected object partially blocking a lane
• Ethical decision-making: choosing between two harmful outcomes such as swerving and risking passengers versus maintaining course and harming pedestrians
• Adverse weather and ambiguous environments: heavy rain or snow obscuring lane markings and confusing sensor inputs

This is not unique to autonomous driving. Many large-scale engineering and innovation projects face similar asymptotic challenges. Early progress creates momentum and optimism, but the final stretch reveals hidden layers of complexity that demand disproportionate effort, time, and resources.

From a project management perspective, this is where traditional planning often breaks down. Linear assumptions fail. Marginal gains become exponentially expensive. And the definition of “done” becomes increasingly ambiguous.

4. The Downfall: when optimism meets reality

Between roughly 2015 and 2020, the autonomous vehicle space experienced a surge of optimism. Capital flowed freely, timelines were aggressive, and the narrative was clear: full autonomy was imminent. Reality, however, had other plans.

From 2021 onwards, a noticeable shift occurred. Several major players scaled back or exited the race altogether:

• Ford and Volkswagen shut down their joint autonomous driving initiative (Argo AI).
• Lyft divested its autonomous unit.
• Apple discontinued its long-running “Project Titan” in 2024.
• Even Uber had previously stepped back after safety incidents.

In total, the industry has seen tens of billions of dollars invested with limited commercial return. Several factors contributed to this downturn:

• Technical challenges proving harder than anticipated.
• High burn rates with uncertain timelines.
• Safety incidents eroding public trust.
• Regulatory uncertainty.

The result is a classic pattern seen in innovation cycles: inflated expectations followed by a trough of disillusionment.

5. The Deeper Insight: when knowledge outpaces technology

At a conceptual level, the autonomous vehicle journey highlights a broader strategic misalignment, one that can be framed through a simple but powerful lens: the relationship between knowledge and technological readiness.

This dynamic is often captured in frameworks that map what we know versus what we can reliably build, similar in spirit to technology maturity models such as Technology Readiness Levels (TRLs), where theoretical understanding can significantly outpace real-world deployment capability.

In this case, the industry possessed:

• Strong theoretical understanding.
• Advanced algorithms.
• Significant computational progress.

However, the supporting technology stack, from sensor reliability to real-world generalization, has not matured at the same pace, thereby creating an imbalance. We know what to build, but we cannot yet build it reliably at scale. The consequences are predictable:

• Overpromising based on theoretical feasibility.
• Underestimating integration complexity.
• Erosion of stakeholder trust over time.

For project leaders, this serves as a powerful reminder: feasibility is multidimensional. Technical possibility does not equal operational readiness.

Closing Reflection

The story of autonomous vehicles is not one of failure but of premature certainty. It is a case study in how ambitious projects evolve:

• They begin with vision and momentum.
• They accelerate through early wins.
• They stall at the edges of complexity.
• And ultimately, they either adapt or pause until the ecosystem catches up.

Autonomous driving may yet become a reality. But its timeline will be dictated not by ambition, nor by investment, but by the resolution of that final, stubborn 10%.

And in project management, as in engineering, that last 10% is where the real work begins.