From RPA bots to AI agents: automating without agent-washing
Gartner warns that many products sold as AI agents are rebranded bots. How to choose between RPA, assistants and agents, and govern agents once live.
The short answer: Use classic automation (RPA and workflow tools) for stable, rule-based tasks, assistants for finding and summarising information, and AI agents only where a process genuinely needs judgement across several steps. Gartner's guidance (June 2025) follows the same split: organisations "can start by using AI agents when decisions are needed, automation for routine workflows and assistants for simple retrieval" (Gartner). The hard part is telling a real agent from a rebadged bot, and governing it once it can act.
This guide is for operations, finance and IT leaders in the UAE and GCC who already run automation and are now being sold "agentic" upgrades.
Why this matters now
Interest is running well ahead of deployment. In Gartner's 2026 CIO and Technology Executive Survey (reported April 2026), only 17% of organisations had deployed AI agents, while more than 60% expected to do so within two years (Gartner).
Results are uneven. In McKinsey's 2026 State of AI survey, 40% of respondents from organisations with annual revenues of more than US$1 billion report scaling AI agents, up from 27% a year earlier, while the share at smaller organisations stayed flat at 22%. Yet only 37% of respondents attribute any EBIT impact to AI, about the same as the year before (McKinsey).
In June 2025 Gartner also predicted that over 40% of agentic AI projects will be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls (Gartner).
What "agent-washing" looks like
In the same June 2025 release, Gartner uses the term "agent washing" for rebranding existing products, such as AI assistants, RPA and chatbots, without substantial agentic capabilities. It estimates that only about 130 of the thousands of agentic AI vendors are real (Gartner).
That does not make RPA obsolete. A bot that posts journal entries from a fixed template, every night, with no exceptions, is doing exactly what it should. The problem comes when a scripted bot is sold, priced and governed as if it could reason, or when a genuinely autonomous agent is governed as loosely as a bot.
Questions to ask any vendor
- What does it decide? Ask for a specific example where the system chooses between options without a predefined rule.
- What can it touch? Which systems can it read, and which can it write to? Under whose identity?
- How does it fail? What happens when data is missing or contradictory? Who is alerted?
- How is it measured? What accuracy, cost per task and exception rate should you expect, and how will you see them?
- What does it cost to run? Usage-based pricing can change the business case at scale.
If the answers describe fixed rules and screen steps, you are buying automation. That may be fine, but price and govern it accordingly.
A simple decision rule
| Work pattern | Best fit | Example in finance or operations |
|---|---|---|
| Stable rules, structured data, high volume | RPA or workflow automation | Moving approved invoices between systems |
| Finding, reading or summarising information | Assistant | Answering policy questions from a knowledge base |
| Multi-step work needing judgement, with clear limits | AI agent with human oversight | Investigating a payment mismatch and proposing a fix |
| High risk, low volume, unclear rules | People, supported by tools | Approving a write-off |
Most real processes mix these. A typical accounts payable flow might use document extraction, rules-based matching, an agent to investigate exceptions, and a person to approve anything unusual.
Redesign the work first
The strongest pattern in the research is that value comes from changing the process, not from adding AI on top of it. In McKinsey's 2026 survey, nearly three-quarters of AI high performers report fundamentally redesigning workflows because of their AI use, compared with one-quarter of other respondents (McKinsey). Gartner makes a similar point: integrating agents into legacy systems can be complex, and rethinking workflows from the ground up is often the better path (Gartner).
In practice, map the current process, including the exceptions people handle by email and spreadsheet. Those exceptions are usually where an agent adds value, and where a bot breaks.
Govern by autonomy level
Not every agent needs the same controls. In May 2026 Gartner predicted that by 2027, 40% of enterprises will demote or decommission autonomous AI agents because of governance gaps found only after production incidents. It recommends classifying agents by autonomy level (Gartner):
- Observe: read-only access, output visible only to the user. Baseline controls such as scoped data access, authentication and logging.
- Advise: drafts and recommendations that people review and act on. Add accuracy testing and user training on how far to rely on the output.
- Act with approval: the agent can change data or send messages after explicit human approval. Add clear approval workflows, audit trails and agent-specific incident response.
- Act autonomously: the agent acts within guardrails while people review exceptions and outcomes. Add continuous monitoring, rollback and circuit breakers that stop the agent when thresholds are breached.
Start most finance and operations use cases at levels 1 to 3. Move up only when you have evidence that the agent is reliable for that task.
Watch the running costs
AI is not free to operate. About 20% of McKinsey's 2026 respondents say AI-related operating costs, including token costs, have constrained their AI use (McKinsey). Build cost per completed task into the business case, and compare it with the fully loaded cost of the current process, including the existing bot licences you might retire.
A six-step plan for existing RPA estates
- Inventory your bots. List each automation, its owner, volume, failure rate and the exceptions it hands back to people.
- Find the exception hotspots. High exception volumes point to where judgement is needed and where an agent might help.
- Redesign one process completely, on paper first. Decide which steps stay rules-based, which need an assistant or agent, and where people decide.
- Pilot at a low autonomy level. Run the agent in observe or advise mode alongside the current process and measure accuracy and cost.
- Set controls before scaling. Identity, permissions, logging, approval rules and an off switch.
- Retire or refactor bots deliberately. Keep the ones that work. Replace the brittle ones only when the new design has been shown to work on your own data.
Whatever automation platform you run today, the same three questions apply. What does each component decide? What can it change? How will you know when it goes wrong?
The bottom line
RPA and AI agents are not rivals. They are different tools for different kinds of work. Organisations that match the tool to the task, redesign the process around it and govern agents by what they are allowed to do will avoid paying agent prices for bot results, or bot-level controls for agent-level risk.
Sources
- Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027" (25 Jun 2025). https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027
- Gartner, "What the 2026 Hype Cycle for Agentic AI Reveals" (15 Apr 2026). https://www.gartner.com/en/articles/hype-cycle-for-agentic-ai
- Gartner, "Gartner Says Applying Uniform Governance Across AI Agents Will Lead to Enterprise AI Agent Failure" (26 May 2026). https://www.gartner.com/en/newsroom/press-releases/2026-05-26-gartner-says-applying-uniform-governance-across-ai-agents-will-lead-to-enterprise-ai-agent-failure
- McKinsey, "The state of AI in 2026: On the road to ROI" (25 Aug 2026). https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
