The agent executes. The human decides.
Which parts of HR work can you hand to an AI agent, and which must stay with a person who is accountable for them?
Saudi Arabia’s Council of Ministers has named 2026 the Year of AI. In HR, the question is no longer whether to use AI. It is which part of our work we hand to an AI agent, and which part must stay with a person who is accountable for it.
Getting the answer wrong is costly in both directions. If your HR department is too cautious, its team stays buried in repetitive work that an agent could finish in minutes. If it is too trusting, it hands an automated system decisions that affect people’s livelihoods and rights.
What is the difference between an AI assistant and an AI agent?
An AI assistant, such as a chatbot, answers your question and stops. An agent is given a goal. It plans its steps, uses tools and systems, and moves information between them until the task is done, within the permissions you give it.
A simple example: an employee asks about their leave balance. The assistant explains the leave policy. The agent checks the balance in the HR system, prepares the leave request, sends it to the manager for approval, and updates the record once it is approved.
The idea is reaching government services too. In September 2026, Saudi Arabia’s Ministry of Human Resources and Social Development said it plans to launch a “Digital Inclusion Agent”, which it describes as an AI agent that gives beneficiaries interactive access to a service or in-kind support without having to log in to online platforms.
Where do AI agents really save HR time?
The clearest return comes from work that is frequent, rule-based and easy to correct when something goes wrong. In HR, five areas stand out:
- Repeat employee questions: leave, official letters and benefits. The agent answers only from approved policies and passes everything else to a specialist.
- New hire onboarding: requesting the account and the laptop, scheduling the first days, collecting documents, and reminding each party of its part on time.
- Interview scheduling: coordinating times between candidates and panel members, and chasing missing documents.
- First drafts: job descriptions, letters and monthly report summaries, always reviewed by a specialist before approval.
- Dates and entitlements: contract expiries and renewals (and their recording on Qiwa in Saudi Arabia) and the end of probation periods, with an alert to the person responsible well in advance.
In this work, the agent makes no decision that affects anyone. It carries out known steps that a specialist used to spend hours on every week.
Where must a human stay in charge?
The rule I propose is simple: the agent executes; the human decides. A person keeps the decision on anything that affects an employee’s livelihood, rights or career:
- Accepting or rejecting a candidate
- Performance ratings and promotion
- Pay and bonuses
- Warnings, disciplinary action and termination
- Compassionate cases and exceptions to policy
For these decisions, the agent can assemble the file, summarize the data and prepare the draft. But the decision and the signature belong to a named person.
This is not only an ethical position. The AI Ethics Principles issued by the Saudi Data and AI Authority (SDAIA) in September 2023 list humanity, transparency and explainability, and accountability and responsibility among their seven principles, and they apply to public and private entities in Saudi Arabia. The Personal Data Protection Law also applies to automated processing that uses employees’ personal data.
In my view, the biggest risk is not that the agent makes a mistake. Every system does. The risk is that a mistake happens and nobody in the department knows who owns the decision the agent carried out.
A fictional example: Al-Rashed Industries
Al-Rashed Industries is a fictional company we use in our training programs: a diversified Saudi group. Its shared services team receives hundreds of questions about leave and official letters every month. So the team built an agent that answers only from the approved policy manual, and drafts routine letters and sends them for approval before they are issued.
But the team did not give the agent authority to decide on exceptional leave requests. The agent collects the request, the leave balance and the relevant policy clause, then passes the file to the manager with a three-line summary. The decision stays with the manager. The team tracks three things: time to reply, the share of correct referrals, and the number of letters changed after review.
A three-question test before you let an agent act
- Can a mistake be reversed easily and without harm?
- Does the action affect someone’s livelihood or rights?
- If an employee asks “why?”, can we explain what the agent did, and on what basis?
If the answer is “no” to the first or third question, or “yes” to the second, put a human approval point before the action, and keep an audit log that records every step.
Why do AI agent projects stall?
Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, because of escalating costs, unclear business value or inadequate risk controls. It also estimates that only about 130 of the thousands of vendors marketing “agents” offer the real thing, and calls the rebranding of ordinary products as agents “agent washing”.
On the other hand, Gartner expects at least 15% of day-to-day work decisions to be made autonomously through agentic AI by 2028, up from 0% in 2024.
The lesson for HR is clear: start with one process and an outcome you measure, not with a tool you buy.
Where to start: four steps in 30 days
- Pick one process that is frequent and low risk, such as leave questions or interview scheduling.
- Map it step by step: the inputs, the decision rules, the systems it passes through, and the human approval points.
- Set the agent’s permissions: what it reads, what it writes and what it never touches. Turn on an audit log for every action.
- Measure the result for 30 days: time saved, error rate and employee satisfaction. Then decide: scale, adjust or stop.
If you want your team to build an agent that works, with human approval points, an audit log and no coding, that is what Agentic AI in Practice (2 days) is designed for. Next English cohort: 1 to 2 March 2027, live online from 09:00 to 13:00 Riyadh time, or onsite in Dubai. See all 2027 dates.
To govern agents and their risks, see Responsible AI: Governance, Ethics & Risk. If your organization is in Saudi Arabia, see our AI and HR training for organizations in Saudi Arabia.
Before you start, find out where your HR department stands: the free AI readiness assessment, 15 questions in 4 minutes.
Frequently asked questions
Will AI agents replace HR staff?
They take over part of the repetitive work, not the roles themselves. HR specialists keep what needs judgment, discretion and accountability: decisions that affect people, exceptional cases and building trust. The skills in demand are shifting toward process design and supervising agents.
Do you need coding skills to build an AI agent?
Not in many cases. No-code tools such as Make or Zapier let you build multi-step agents. The most important skill is to map the process, the decision rules and the approval points clearly.
What should organizations in Saudi Arabia consider from a regulatory point of view?
At a minimum, the Personal Data Protection Law, because the agent processes employee data, and SDAIA’s AI Ethics Principles. This is not legal advice; check with your organization’s legal counsel before running any agent on real data.
Sources
- Council of Ministers names 2026 the Year of AI: Maaal, 10 March 2026 (Arabic)
- Ministry of Human Resources and Social Development, Digital Inclusion Agent: Al Eqtisadiah, 4 September 2026 (Arabic)
- SDAIA AI Ethics Principles (September 2023) and the Personal Data Protection Law: Herbert Smith Freehills Kramer, AI Tracker: Saudi Arabia
- Gartner, 25 June 2025: Over 40% of Agentic AI Projects Will Be Canceled by End of 2027