In September 2016, Uber began asking drivers to snap a selfie before they could accept fares. The feature—dubbed Real-Time ID Check—wasn’t just a safety tweak; it was a working proof of Microsoft’s grand AI ambitions. At a time when rivals chased headline-grabbing game victories, Microsoft leaned heavily into cloud-powered, enterprise-ready intelligence. Eight years later, the selfie check stands as a small but telling milestone in a winding journey that shifted from Cortana’s consumer dreams to the Copilot era’s enterprise dominance.

Microsoft’s 2016 AI Blueprint

At the 2016 Microsoft Ignite conference, CEO Satya Nadella laid out a vision that defied the industry’s obsession with gaming benchmarks. “We are not pursuing AI to beat humans at games,” Nadella said, instead framing AI as a fundamental tool to transform everyday work. The strategy hinged on four pillars: agents (Cortana), applications (Office, Dynamics), services (Cognitive Services, Bot Framework), and infrastructure (Azure plus specialized hardware).

Microsoft moved quickly to open its assistant to developers. The Cortana Skills Kit and Devices SDK let third-party apps integrate with Cortana, turning the assistant into a platform rather than a standalone service. The company touted 145 million monthly active users at the time—a figure that, while later disputed, underscored the scale of ambition.

On the infrastructure front, Project Catapult deployed FPGA accelerators across Azure data centers. These reprogrammable chips were designed to speed neural network workloads, making AI services faster and more cost-effective. “We’re betting that FPGAs will become the essential ingredient of a future cloud,” a Wired article noted, capturing Microsoft’s bet on non-traditional silicon.

Partnerships formed another key plank. Adobe named Azure its preferred cloud platform, tying Creative Cloud, Marketing Cloud, and Document Cloud to Microsoft’s infrastructure. The collaboration promised integrated workflows and shared data models, giving Azure a foothold in creative and marketing industries.

The Uber Selfie: AI in Action

The Uber driver selfie feature crystallized Microsoft’s approach. Using Microsoft Cognitive Services, Uber’s Real-Time ID Check compared a live selfie against a stored profile photo. If the images didn’t match, the driver’s account was temporarily locked, reducing the risk of driver fraud and enhancing rider safety.

TechCrunch reported that during a pilot, over 99% of drivers passed verification. Mismatches usually stemmed from unclear profile pictures rather than impersonation. Uber later expanded the feature internationally, demonstrating how cloud AI could handle sensitive, real-time identity checks at scale.

The system relied on Microsoft’s Face API, which performed facial matching without storing raw biometric data beyond what local laws allowed. Yet the feature also raised privacy questions. Uber’s data retention practices, later clarified, showed that photos were kept to track verification outcomes—a standard logging practice, but one that demands clear privacy disclosures.

Strengths, Gaps, and the Long Game

Microsoft’s 2016 playbook had clear strengths. A platform-first approach with Cortana Skills and the Bot Framework invited developer innovation. Infrastructure bets on FPGAs and GPUs gave Azure a performance edge as AI workloads exploded. Enterprise partnerships like Adobe and Uber created real revenue streams and reference cases.

But execution gaps emerged. Cortana never became a dominant consumer assistant. Despite early user numbers, Microsoft quietly retired the standalone voice assistant in 2023, shifting focus to Microsoft 365 Copilot and Bing Chat. The Cortana Skills Kit, while technically sound, lacked the sustained product commitment developers needed.

Independent verification shows that while most of Microsoft’s 2016 announcements are well-documented, the subsequent narrative of Cortana’s user base veered into misinterpretation. A Mashdigi roundup, for instance, cited a 1.33 million user figure that conflicts with Microsoft’s contemporaneous blog post claiming 145 million. This gap illustrates how quickly AI metrics can become distorted without rigorous sourcing.

Where the Vision Stumbled

Cortana’s consumer presence, once pegged at hundreds of millions of users, never matched Alexa or Google Assistant. By 2023, Microsoft officially retired Cortana as a standalone voice assistant in Windows. Support documents directed users toward Microsoft 365 and Copilot alternatives. The Cortana brand, once central to the “agents” pillar, was folded into narrower enterprise features. The platform play was strategically sound—opening APIs and inviting developers—but sustained consumer adoption required a compelling, continuously updated product that Microsoft never quite delivered.

The Pivot to Copilot and Enterprise AI

While Cortana faded, the infrastructure bets paid off spectacularly. Azure’s GPU and FPGA resources positioned Microsoft as a cloud leader for large model training and inference. The platform that once powered driver selfies now underpins Azure OpenAI Service and Microsoft 365 Copilot, embedding generative AI into Word, Excel, Teams, and Outlook.

Nadella’s 2016 line about AI not being for games proved prescient in a different way. As NVIDIA’s Jensen Huang later argued, AI became a new computing platform, not a novelty feature. Microsoft’s data center investments—including a deepening partnership with OpenAI—transformed Azure into the backbone of enterprise AI workloads. Copilot, not Cortana, became the face of Microsoft’s AI in the 2020s.

The shift reflects a hard-earned lesson: developer platforms need not just openness but stable, long-term product focus. Cortana’s Skills Kit, while innovative, never achieved the critical mass that would lock in developers. In contrast, Copilot integrates directly into tools millions of businesses already pay for, creating immediate value and stickiness.

Ethical Tightropes and Governance

The Uber selfie example also exposed ethical fault lines. Facial recognition systems have well-documented biases, often misidentifying people of color and women at higher rates. Deploying such systems for identity verification without robust human review, diverse training data, and clear appeal processes risks harmful outcomes.

Microsoft’s own responsible AI guidelines evolved over time, but the 2016 partnerships predated widespread industry guardrails. Today, any organization using biometric or identity AI must audit for bias, ensure explainability, and give users a path to contest automated decisions. The lesson from Uber’s rollout: over 99% verification sounds reassuring, but even a small error rate can impact thousands of drivers when scaled globally.

Platform concentration raises another concern. Adobe’s deep ties to Azure, Microsoft’s control over Copilot, and the company’s heavy influence in AI tooling create dependencies. For IT buyers, that means evaluating exit strategies, data portability, and multi-cloud options before committing to an AI stack.

What Windows Users and IT Leaders Should Take Away

The 2016-2024 arc offers concrete takeaways for anyone managing Windows environments or building on Microsoft’s AI services:

  • Design for change. Platform bets can shift. Architect systems with modularity to swap components if a service gets deprioritized.
  • Scrutinize metrics. Whether it’s “active users” or model accuracy, demand clear definitions. Cortana’s user-count confusion was a cautionary tale.
  • Govern responsibly. Any AI system that makes decisions about people—identity checks, content moderation, meeting summarization—needs bias testing, human oversight, and incident response plans.
  • Expect AI to be embedded, not standalone. Copilot in Windows 11, Microsoft 365, and Azure signals a future where AI functions as a layer inside existing tools, not a separate assistant app.
  • Watch the cost curve. Specialized hardware accelerates AI but can inflate bills. Pilot workloads and benchmark CPU vs. GPU vs. FPGA to find the break-even point.

For developers, the 2016 Cortana Skills Kit was a reminder that APIs alone don’t build ecosystems. Stability, monetization, and tooling matter just as much as reach.

From a 2016 Selfie to a 2024 Copilot

In September 2016, a driver’s quick selfie verified more than their identity—it validated a corporate strategy. Microsoft’s insistence that AI wasn’t a gaming sideshow led to real, if uneven, progress. The company’s early investments in cloud infrastructure, specialized silicon, and enterprise partnerships laid the groundwork for today’s Copilot-fueled productivity tools.

But the journey also bruised. Cortana’s consumer retreat, the noise around user numbers, and the ethical questions raised by facial recognition all stand as reminders that platform announcements are the easy part. Turning vision into trustworthy, adopted, and sustainable products requires constant refinement.

The Uber selfie check now feels like a historic footnote. Yet in its blend of cloud APIs, real-time intelligence, and operational safety, it was a seed. The full bloom—Copilot embedded across the Microsoft ecosystem—took nearly a decade, a few pivots, and a willingness to kill darlings like Cortana. As Windows users and enterprises step into the Copilot era, the 2016 promise finally finds its footing: AI that isn’t for games, but for getting work done.