Agents for leave and time tracking, an assistant on your company's data, MCP connectors and a daily summary for managers. Decisions stay deterministic, and a person confirms every action.
Translated from Romanian. This article covers Romanian labour and tax law; the Romanian version is canonical and is updated first.
For many HR departments, the first step was digitalisation: paper attendance registers, personnel files and spreadsheets were replaced by a platform where time tracking, leave and documents live in one place. The benefits show quickly — fewer lost documents, information found in seconds, time records that no longer have to be reconstructed at the end of the month. The next step is a natural one: the repetitive tasks that remain — checking time records, approving leave requests, answering the same questions — can be automated. On one condition: without anyone losing control over the decisions.
The discussion about AI in HR usually gets stuck in one place: a language model can phrase a convincing but wrong answer. For questions about schedules or leave, that’s annoying; for a decision that affects an employee, it’s unacceptable. We explained the difference between generative AI, classification and rule-based systems at length in our article on AI in HR.
That’s why the AI agents in Workly start from a simple rule: decisions are deterministic, and the language model is only the language layer. The model understands the question and drafts the answer. The decision belongs to the rule you configured — the same input always gives the same result, and the result can be explained step by step.
You configure the company’s policies — minimum notice, how many people can be away from a team at the same time, available balance, blocked periods. When a request is submitted, the agent applies the rule and approves or rejects it on the spot. It isn’t an estimate: it’s your policy, applied consistently to every employee. Each decision keeps the rule applied and its parameters, and the employee can contest a rejection — the request then goes to a person.
Every night, the agent scans time records against six anomaly rules, with thresholds you set, and some situations — for example, clocking in on a day of approved leave — are flagged immediately. What it finds are suggestions to check: they don’t stop anyone clocking in and don’t penalise anyone.
You ask in natural language — who is on leave next week, how much overtime the team did this month — and the answer is built from the company’s real data, through tools with permissions. The security layer is the tools, not the instruction given to the model: everyone sees only what they’re allowed to see, and sensitive data is masked before it leaves the server. When the assistant can do something for you — for example, prepare an email — it only proposes it: the action runs only after you confirm it explicitly, and an email goes out with exactly the text you saw and could edit.
If you already use an AI assistant that supports the MCP protocol, you can connect it to Workly. The connector is personal: each employee generates their own access, and the assistant sees strictly what that person sees in the app, read-only. Access expires after a period of inactivity and is closed automatically when someone leaves the company, and the administrator has a kill switch for the whole company or for a single person.
Every morning, each manager receives by email and as a notification yesterday’s activity for their team: hours worked, absences, today’s leave, the requests waiting for them and the time tracking agent’s flags. The figures come straight from the database; the language model only drafts the opening sentence. When there’s nothing to report, you receive nothing.
Where HR is a single person who also does other things, the agents take over routine checks: leave requests that follow the rules are approved automatically, and problematic time records stand out on their own. The policy applies the same way to everyone, with no “tired” exceptions, and switching it on doesn’t require a technical project.
As the structure becomes more complex, policies can be differentiated, and the HR team is freed from repetitive approvals. Anomalies — overtime piling up, repeated absences, missing time records — are spotted before they become a compliance risk, and managers get a daily picture of their team without asking for it.
With hundreds of employees, manually checking every day of time records is no longer realistic. Rules applied uniformly remove the inevitable variation between people, MCP connectors link the data to the AI tools the company already uses, and HR gets its time back for the strategic side.
The European regulation on artificial intelligence requires systems used in employee management to be explainable and overseen by people. In Workly:
The AI agents are an optional module, enabled per company, and each agent is switched on separately. Only the conversational assistant and the opening sentence of the daily summary use a language model; the leave and time tracking agents, like Modi — the internal assistant — work without one. The details of each agent and its limits are on the AI Agents module page, and the easiest way is to see the agents working on your company’s rules, in a demo. If you first want to check which claims about AI in HR hold up, the article Four myths about AI in HR runs each claim through the law and the data.
Informational article about Workly’s features at the date of publication. The availability of each feature depends on the modules active in the company’s account.
Workly newsletter
Get Workly news (new features and modules), the legislative changes that matter for HR and payroll, new document templates and the practical guides we publish on the blog, by email.
We'll show you how time tracking, payroll and the rest of the modules work for your company, in a short demo.
Request a demo