Agentic AI in HRIS: What Pakistani HR Teams Should Know

Agentic AI in an HRIS is software that plans and carries out a multi-step HR task on its own, inside rules you set, and stops for a human at the decisions that matter. An AI assistant answers a question. An agent finishes the job: it checks a leave request against policy, updates the balance, tells the manager and writes down what it did.

Most Pakistani HR teams do not need that today. Most do need to understand it, because the vendors selling it are loud and the evidence is thin. Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027. This guide shows where an agent earns its place inside an HRIS, where it does not, and what to fix first.

An assistant waits for a question and replies. An agent receives a goal, breaks it into steps, acts in the system and adapts when something changes. In HR terms, the first tells an employee how many casual leaves are left. The second also routes the request, checks the team calendar, updates the record and flags the manager only if the request breaks a rule.

The label matters less than the behaviour, so test behaviour. Gartner warns about agent washing, the rebranding of assistants, robotic process automation and chatbots as agents. It estimates only about 130 of the thousands of agentic AI vendors are real. Anushree Verma, Senior Director Analyst at Gartner, said many use cases positioned as agentic today do not require agentic implementations (Gartner, June 2025).

For you, that means a demo labelled agentic may be a chatbot with a new name. Ask the vendor to run one multi-step task on your own sample data and watch every step.

Pick tasks that follow written rules, repeat every month and cause little harm when something goes wrong. Leave routing under policy fits. So do attendance exceptions, onboarding checklists, payroll input checks before a run, probation and contract end reminders, and statutory deadline alerts for EOBI and PESSI. Our notes on attendance tracking beyond biometric machines and on leave approval costs show how much of this work is rule-following today.

Then decide how much freedom the agent gets. Think in three levels, and move a task up only after it has run cleanly at the level below.

The interest is real and the results are thin. McKinsey's 2025 survey of 1,993 respondents in 105 countries found 23% of organisations scaling an agentic AI system somewhere and another 39% experimenting. Yet no more than 10% were scaling agents in any single business function, and only 39% reported any EBIT impact from AI at all (McKinsey, November 2025).

The same survey names what separates the 6% of high performers: they were nearly three times as likely to have fundamentally redesigned individual workflows. That is the lesson for a Pakistani HR team. An agent follows a process. If your leave approvals live in WhatsApp messages and your attendance reconciliation lives in one person's Excel file, there is no process to hand over. The agent either does nothing or automates the mess.

I architected the HR workflows in Workflow Engine V1 and run HR at Bitsbuffer, so my read is simple: define the workflow first, then decide what to automate. We wrote the longer argument in build the workflow before the agent and in the workflow-first approach.

Keep a person on every decision that changes someone's pay, job or record of conduct. That covers final payroll approval, hiring and rejection, salary changes, disciplinary action, termination and any edit to a statutory deduction. An agent can prepare the evidence for these. It should not make the call.

Regulators in other markets already treat this area as high risk. The EU AI Act classes AI systems used in HR and employment as high-risk, and a provisional agreement reported in June 2026 would move the application date for those systems to 2 December 2027 (Sidley, June 2026). That law does not bind a Pakistani employer. We cite it as an example of where careful governance is heading. Our guide to ethical AI governance in an HRIS turns the principle into a checklist.

Work through this order. Each step is cheap, and skipping one is how pilots end up cancelled.

1) Pick three workflows with the most monthly repetition, such as leave, attendance exceptions and onboarding. 2) Write the rules for each in plain sentences, including who approves what and what counts as an exception. 3) Clean the employee records those rules read from. 4) Start every task at level 1 or 2 for one full payroll cycle. 5) Define stop conditions: the agent halts and asks a person when data is missing or a rule conflicts. 6) Review the action log weekly and promote a task to level 3 only when exceptions stay low.

Workflow Engine is Bitsbuffer's HRMS for Pakistani companies, and we suggest holding it to this list too. Bring your three workflows to a demo and we will say plainly which of them it covers today and which it does not.

Do not expect a Pakistan benchmark. We found no Pakistan-specific study of agentic AI in HR, so every figure in this guide is global. Treat them as a caution, not a forecast for your company.

Gartner predicts at least 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028, up from 0% in 2024. Read that as a direction, not a promise. Your payroll will still need a person who understands why a number is right, and so will your employees.