Skill Spotting or Surveillance? AI's New Role in Evaluating Work

SAN FRANCISCO, CA – October 28, 2026 — For generations, the trajectory of a professional life has been dictated by proxies. We rely on resumes, self-evaluations, university degrees, and the often-subjective memories of middle managers to determine who gets promoted, who gets reskilled, and who gets let go. According to new industry data, it is a system built on a foundation of whispers: six in ten managers admit they make critical talent decisions based entirely on guesswork.

Today, a new technological framework threatens to upend this fragile system. CodeSignal, a platform historically known for technical assessments, has announced the general availability of its Agentic Skills Intelligence platform. The premise is as ambitious as it is controversial: rather than asking employees what they can do, the software deploys specialized AI agents to continuously evaluate the work they are already doing.

By analyzing digital artifacts from everyday enterprise systems—including Salesforce, GitHub, Zoom, Cursor, and Google Workspace—the platform builds a living, validated skills profile for every worker. It is a profound shift from the static performance reviews of the past, but as we integrate AI deeper into the fabric of our daily labor, we must ask: at what point does continuous evaluation cross the line into pervasive surveillance?

Beyond the Resume: The Era of Continuous Validation

For years, human resources departments have relied on platforms like Workday Skills Cloud or Eightfold AI to map workforce capabilities. These legacy systems largely depend on inference—scraping keywords from LinkedIn profiles or analyzing self-reported competencies to guess an employee's potential.

The new agentic approach discards inference in favor of direct observation. The software acts as a silent observer, scoring raw work output against structured rubrics grounded in industrial-organizational psychology. If a sales representative delivers a pitch on Zoom, the system evaluates their communication and product knowledge. If an engineer writes code in GitHub or Cursor, the technology assesses their technical proficiency in real-time.

"Every leader is being asked hard questions about their business right now. Is our AI investment paying off? Do we have the skills to hit our goals? Who's ready to lead the next big project? Too often the answer is a guess based on weak data signals," said Tigran Sloyan, CEO and co-founder of CodeSignal. "We built Agentic Skills Intelligence because the people decisions that matter most deserve real data. Now leaders can practice what I call skillspotting at scale: seeing what their people can actually do, based on the work they're already doing."

When a skills gap is identified, the system doesn't just flag it; it attempts to fix it. An integrated AI tutor named Cosmo transforms these detected deficiencies into contextualized, one-on-one learning practices grounded in the employee's actual daily workflow, theoretically closing the loop between assessment and development.

The Multi-Million Dollar AI Blind Spot

The corporate appetite for this level of visibility is driven by a looming financial crisis in enterprise technology. Over the past year, 84% of organizations expanded their generative AI software investments, and 79% increased AI training. Yet, a staggering 79% of these same organizations cannot measure the return on investment of that spending.

Recent studies from institutions like MIT Sloan and KPMG paint a stark picture: while a majority of companies report anecdotal productivity gains from artificial intelligence, over 80% report no tangible enterprise-level profit-and-loss impact. Companies are buying seats for Claude and ChatGPT, but they have no objective way to measure if their employees are actually using these tools effectively to drive business outcomes.

"In the AI era, leaders need to understand not just what skills they can infer their people have, but what people can demonstrate," noted Josh Bersin, global industry analyst and CEO of The Josh Bersin Company. "CodeSignal adds evidence of proficiency, how well people apply skills in their work, and connects it to personalized development. That helps leaders make better decisions while giving employees a clear path to grow."

By establishing a baseline of employee capabilities and continuously tracking output, executives hope to finally quantify the impact of their technology investments. If a CEO spends millions on a new generative suite, agentic intelligence promises to show exactly who is leveraging the tools to produce better work and who is falling behind.

The Thin Line Between Evaluation and Surveillance

However, the transition to a continuously monitored workforce carries profound ethical and psychological implications. The passive ingestion of communication and work tools strikes at the heart of workplace privacy, raising alarms among labor advocates and legal experts.

Workplace privacy researchers have long warned about the dangers of scope creep. A system introduced under the guise of personalized learning and skill development can easily be weaponized during economic downturns to justify layoffs or suppress wages. When every Slack message, Zoom call, and line of code is algorithmically graded, the psychological safety required for creative, high-stakes knowledge work is fundamentally altered.

"There is a massive difference between taking a standardized test and having an algorithm silently grade your every interaction," noted one enterprise labor attorney who advises Fortune 500 companies on employee monitoring compliance. "Without radical transparency and ironclad informed consent, these systems risk destroying the trust between employer and employee. Workers need to know exactly what is being measured, who has access to that data, and how it will be used."

Furthermore, the reliance on algorithmic evaluation introduces the risk of systemic bias. While the creators claim to utilize structured rubrics designed by industrial-organizational psychologists to mitigate prejudice, the reality of machine learning is often more complex. Models trained on historical enterprise data can inadvertently learn and perpetuate the very human biases they are meant to replace. Ensuring construct validity—proving that the AI is actually measuring skill and not just a worker's ability to game the algorithm—will require rigorous, continuous auditing.

A Standardized Future for Human Capital

Despite these concerns, the technological momentum toward agentic evaluation appears unstoppable. The new platform is built on the Model Context Protocol (MCP), an open-source standard introduced by Anthropic that allows AI applications to seamlessly connect with external data sources.

This standardization means that managers don't even need to log into a separate human resources dashboard to check on their team's skills. They can simply ask their preferred AI assistant to pull a real-time skills report or identify the best internal candidate for a critical new project. The data flows seamlessly from the employee's keyboard, through the evaluation engine, and directly into the manager's chatbot.

As we navigate this new frontier, the fundamental social contract of the workplace is being rewritten. The promise of objective, data-driven talent management is undeniably appealing, offering a potential end to the nepotism, bias, and guesswork that have long plagued corporate advancement. Yet, in our pursuit of perfect efficiency and measurable returns, we must carefully consider the human cost of a world where our every professional keystroke is weighed, measured, and judged by an invisible machine.