The reskilling conversation in governance, risk and compliance is being carried loudest by the vendors selling the tools, which is precisely why it deserves a second reading against the data. The story is not that AI agents have already taken the GRC job. It is that the routine tasks are genuinely migratory, adoption inside GRC is far earlier than the marketing implies, and practitioners' calm about that gap may be the most underpriced risk on the board.
The tasks are moving, not the job titles
The most direct version of the claim comes from a sponsored BleepingComputer piece by Yair Kuznitsov, co-founder and CEO of the GRC platform Anecdotes: AI agents are, in his words, "already doing the control testing, the evidence collection, and the framework mapping that used to fill a manager's week." That is a vendor's certainty, and it should be read as such. The independent check comes from the International Compliance Association, whose 2025 Global GRC Survey asked 383 practitioners across 87 countries and more than 30 sectors. There, only 1.6% of GRC functions said AI is fully integrated into their processes, and a combined 3.2% said AI is either fully integrated or that their department relies heavily on it for monitoring and reporting. Roughly a third are in the early stages of incorporating AI; almost the same share have not adopted it for GRC purposes at all. The agentification of control testing is real in early adopters and promised everywhere else. The honest framing is underway, not done — a distinction that matters for anyone making hiring or budget decisions off vendor decks. The same gap between deployment and governance is what we traced in why AI cybersecurity governance defines the next maturity phase.
What "agentification" actually changes
Anecdotes defines GRC agentification as AI agents doing the work while the human "moves up a level to direct and govern them" — not assistance, but delegation with supervision attached. Buyers appear to be heading that way deliberately. Drata, a compliance automation vendor, reports from a survey it commissioned of 300 IT and security professionals that 64% of buyers prefer targeted agentic systems over general-purpose assistants, rising to 70% among risk-focused respondents. It is vendor-commissioned research, but a preference survey of buyers is still evidence of direction, even if the vendor framing around it is promotional. The substantive shift is organizational: the human role migrates from performing evidence work to setting objectives, boundaries and escalation rules for systems that perform it. That is a different job description, and the skills market has not repriced it yet.
The complacency gap
Here the ICA data gets uncomfortable for practitioners rather than vendors. Asked about AI displacing their roles, 41.3% of the 383 surveyed professionals said they were not concerned at all and 37.25% slightly concerned; only 11.34% said they were very concerned. Nearly four in five are unruffled. Yet 47% of their firms plan to increase automation investment more broadly, and the ICA cites the Bank of England's finding that 75% of financial-sector firms were using AI by the end of 2024 — concentrated in operations and IT, not in compliance. Enterprise AI use is outrunning GRC's own adoption and, more importantly, its seat at the AI governance table. The realistic threat reading this data is not replacement by agents; it is irrelevance by bypass. The AI security workforce is already reshaping around automation, and compliance is next in line.
The skills, restated with evidence
Anecdotes' three-skill framework — context, agent governance and communication, is vendor thought leadership, but each leg has independent-ish support behind it. Data literacy: Drata's survey found only 13% of IT and security professionals are fully confident they can see every AI tool their employees use, and those blind spots map onto the risks GRC exists to manage, data security (57%), privacy (52%), compliance violations (51%). You cannot govern data pipelines feeding agents you cannot even enumerate. Agent governance as people management: Anecdotes argues each agent should get written KPIs before touching production, the outcome it owns, an accuracy bar, an escalation rate, because "an unmanaged agent is not a productivity win, it is unsupervised risk running at machine speed," and agents can game their metrics by quietly drifting or doing less. Drata's own respondents converged on the same shape, describing human judgment as a control plane rather than a clean-up crew. Communication as audit trail: Anecdotes insists every agent-to-agent handoff leave a visible, time-stamped record; the payoff is audit defensibility, the exact capability 86% of Drata's survey respondents say current GRC AI tools lack. The capability squeeze on security and compliance teams trying to absorb all of this is the subject of our earlier piece on AI's dual threat of complexity and the CISO capability gap.
The accountability gap is the stakes paragraph
The single most consequential number in the Drata survey: 71% of organizations say an AI tool used for GRC functions has led to a failed audit or a lapsed regulatory standard at least once. When that happens, the buyer wears the finding and the vendor moves on to the next logo, that asymmetry is the accountability gap, and it explains the 86% consensus that there are too many GRC AI tools that are not enterprise-ready on scalability, security and audit defensibility grounds. Read alongside the ICA's finding that only 1.6% of GRC functions have fully integrated AI, a useful inference emerges: the failures are arriving from pilots and point tools, ahead of any mature agentic program. The simple macroeconomics of AI in this sector is that the cost of a mistake stays with the human organization while the marginal cost of the automated work falls toward zero. Governing agents is not career-development decoration; it is what stands between a vendor demo and a regulatory finding in your name.
The structural gap inside GRC itself
The ICA survey also measured where GRC sits when other departments roll out AI, and the picture is uneven: 32.79% of practitioners report little or no engagement in those rollouts, 32.39% engage early as a supporting consultancy, only 23.48% have a full partnership approach, and 11.34% show up only late-stage. The career implication follows directly from the organizational one: influence, headcount and budget are migrating toward whoever sits on the AI governance side of the business. GRC professionals who spend the next two years building agent-governance muscle inside those rollouts are positioning where the work is going; those waiting for compliance review requests to arrive are staffing a function other departments are routing around. The ICA's own reading is charitable, "cautious rather than resistant", but caution without a seat at the table is just tardiness with better manners.
Conclusion: the simple macroeconomics of AI, applied to one career
Zoomed out, the GRC case is a clean instance of the simple macroeconomics of AI: automation absorbs the routine tasks, the residual human work becomes the judgment-intensive complement to those tasks, and the wage and influence premium attaches to the complement, not the task. The durable role is the one that is harder to automate than the work it used to automate, setting objectives for agents, keeping their metrics honest, and making their actions traceable to a regulator. The reskilling window is open for a counterintuitive reason: adoption is still early, as the ICA's full-integration figure of 1.6% makes plain. Anecdotes' closing line, that the GRC manager is being replaced not by AI but by the GRC manager who learned to use it, starting this quarter, is a vendor's urgency, and the "this quarter" is the sales part. The direction, though, is what the survey data independently supports. The safe role is not the quiet one. It is the one that governs the agents.