What happens when AI changes the work long before a company changes the job description?
In this episode of AI at Work, I speak with Vishnu Shankar, Chief Data Officer at Draup, about a workforce problem that is easy to miss when the public debate remains fixed on jobs being created or removed. Vishnu argues that the faster and more immediate change is happening inside existing roles. Tasks that once took days can now be completed in minutes, but people must verify outputs, understand faulty systems, recognize changing naming conventions and apply the judgment that a model lacks.
That is why Draup describes many skills as facing partial obsolescence. SQL, data modeling and similar capabilities do not suddenly become worthless. Their place within the role changes. At the same time, new AI requirements are appearing inside customer support, sales, marketing and software engineering roles. Vishnu says Draup's analysis of Fortune 500 employment data found that dedicated AI roles were growing only marginally, while demand for AI skills within existing roles rose far faster. He also describes increasing skill density, with some software engineering positions asking for 19 skills where 14 may previously have been expected.
We discuss why this creates problems for traditional reskilling. Comparing two static lists of skills misses the context in which each skill is used and ignores how the destination role is changing. Vishnu proposes starting with tasks instead. Leaders should identify what AI will automate, what it will support and what continues to require human effort. Only then can they determine which skills a person needs and whether an adjacent career route is realistic.
The conversation also turns to capabilities that rarely appear in workforce systems. Communication, for example, is too broad to assess as a single skill. In an HR role, it may mean conducting a difficult performance conversation, challenging a senior stakeholder or running a calibration meeting where managers defend their ratings. Evidence of those abilities may live in project records, manager feedback, review discussions or systems such as Jira rather than in a conventional employee profile.
For leaders, the governance question is equally important. Vishnu believes workforce transformation should be led by the CHRO, supported by the CIO's technology roadmap and informed by the CFO's budget authority. He calls for an automation register that records what was automated, where the work previously sat and what replaces it. He also argues that headcount planning becomes less informative when AI changes how much one person can produce. A capability portfolio offers a clearer view of what the organization can actually deliver.
We finish with a direct challenge to employers. Training catalogs, completion records and certificates do not prove capability. People need protected time during the working day, practice using real work with real consequences, and a credible route into a named role or family of roles. Work simulations can then test whether employees apply the skill, use the available tools and catch planted errors. Vishnu also explains how unrealistic experience requirements, overloaded job descriptions, slow hiring processes and weak compensation can all be mislabeled as talent shortages.
Is your organization measuring the number of people who completed AI training, or whether those skills are appearing in the work they produce? Share your thoughts and experiences with me.

