8 Things I Learned Building an HR Tool With AI in 3 Months

In late 2025, I sat at my kitchen table at 11pm, staring at a workflow diagram in n8n, trying to work out why my AI agent was returning the wrong jurisdiction for a general protections question.

I had no coding experience from 2013. I had a redundancy, a 14-year career in HR, and a problem I kept watching HR teams get wrong - the moment a workplace issue arrives on someone's desk and nobody knows where to start.

Three months later, HRWise.com.au was live.

I am not a developer. I am an HR practitioner who got deeply uncomfortable and built something anyway. And the experience taught me things I did not expect - about AI, about building, about what HR expertise is actually worth in the age of these tools, and about where the gaps are for every practitioner who is thinking about doing the same.

---

1. You do not need to be a developer. You need to be precise.

The biggest misconception I had going in was that building an AI tool required coding skills. It does not - but it requires something most HR practitioners are not trained to do: write instructions with absolute precision.

AI agents do exactly what you tell them. Not approximately what you mean. Not what you intended. Exactly what the words say. In HR, we are trained to use nuanced, contextual language. "It depends" is a valid answer. "In most cases" is standard practice. These are catastrophic instructions for an AI agent.

The shift I had to make was from HR practitioner to systems designer. Every decision tree in HRWise.com.au had to be mapped completely before the tool could execute it. That discipline - forcing myself to articulate every branch of a decision - made me a better practitioner, not just a better builder.

What this means for you: Before you touch a tool, write out the process you want to automate as if you are explaining it to someone who has never worked in HR. Every step. Every exception. Every "it depends." If you cannot write it, the tool cannot do it.

---

2. The tools have caught up to the non-developer faster than anyone expected.

When I started, I expected to spend most of my time fighting the tools. I did not. The ecosystem around n8n, Supabase, Google Gemini, and Claude has moved faster than I anticipated. n8n added AI Agents as a native feature in late 2025, which meant I was building with capabilities that were months old. Supabase handled my database and vector store without requiring me to write SQL from scratch. Claude handled the reasoning layer. Google Gemini handled some of the document processing.

The honest assessment: the tools are now accessible to a non-developer who is willing to invest four to six weeks of genuine learning. Not dabbling. Not watching YouTube videos. Learning by doing - building something small, breaking it, understanding why it broke, and building it better.

The piece I underestimated was the integration layer. Getting tools to talk to each other cleanly - passing data in the right format, handling errors gracefully, making sure the workflow does not silently fail - took longer than building any individual component.

What this means for you: The barrier is not the tools. The barrier is the patience to learn through failure. If you can tolerate being confused for six weeks, you can build something meaningful in three months.

---

3. Your HR expertise is the product - the AI is the delivery mechanism.

This is the lesson I wish someone had told me on day one.

I spent weeks trying to make the AI smarter. I kept tweaking prompts, adding more context, trying to get the model to produce better answers. The outputs kept being generic. Good, but not expert-level.

The breakthrough came when I stopped trying to make the AI think like an HR expert and started feeding it my thinking. The frameworks I had developed over 14 years. The decision trees I had used in actual cases. The questions I ask when a manager calls me in a panic about a performance issue.

“

The AI is not the expert. You are. The AI is the mechanism that delivers your expertise at scale, at 2am, to someone who cannot afford to call you.

When I rebuilt the triage logic as if I were training a very capable junior HR practitioner - walking through my actual reasoning, not generic HR principles - the quality of outputs jumped immediately.

What this means for you: The value of an AI tool built by an HR practitioner is not the AI. It is the practitioner's thinking, encoded in the system design. That is your competitive advantage. Protect it. Document it. That is your IP.

---

4. Build the smallest possible thing first - and make it genuinely useful.

My original vision for HRWise.com.au was much larger. A full case management system. Automated documentation. Jurisdiction detection. Real-time Fair Work updates. An advice engine covering every pillar of employment law.

I built none of that first.

What I built first was a single triage flow: a manager describes a workplace issue, the tool asks five structured questions, and it returns a risk classification with three practical next steps. That was it. One flow. One use case. One problem solved.

That constraint was the best decision I made. It forced clarity on what the tool was actually for. It gave me something I could test with real users in week six rather than month six. And the feedback from those early users - HR managers, business owners, people in my network willing to try something unfinished - shaped every subsequent build decision.

What this means for you: What is the single most common question you get from managers or HR teams in your network? Build a tool that answers that question, well, every time. That is version one.

---

5. The Australian ER/IR context is harder to automate than you think - and that is your advantage.

I spent significant time working through this. General commercial AI tools - ChatGPT, even Claude in its base form - handle Australian employment law inconsistently. The Fair Work Act is specific. Modern awards are complex. The general protections reverse onus is not well understood by models trained primarily on US and UK employment law content.

This gap is a genuine competitive advantage for an Australian HR practitioner building tools for Australian HR teams.

The solution I landed on was not trying to make the AI know the law - it was building the law into the system design. Every triage path in HRWise.com.au is mapped against the specific legislative provisions. The AI does not reason about the Fair Work Act from first principles; it follows decision logic that I designed based on the Act and confirmed against the FWC's own guidance.

SHRM's 2026 report found that in organisations that have implemented AI, HR professionals use it frequently - 26% weekly, 20% daily, and 9% several times a day. But the vast majority of those tools are built for global markets. The Australian-specific gap - modern awards, the general protections regime, psychosocial hazard obligations, the right to disconnect - is largely unserviced by commercial tools.

What this means for you: If you are thinking about building something for Australian HR teams, your knowledge of the specific local legislative context is not a nice-to-have. It is the entire point of difference.

---

6. Privacy and data handling cannot be an afterthought.

This one I did not think about early enough.

HR data is some of the most sensitive data that exists. Workplace complaint details. Performance records. Health information disclosed in a medical certificate. When you build a tool that processes this data, you are immediately in the territory of the Privacy Act 1988 (Cth) and the Australian Privacy Principles.

The questions I had to answer - and should have answered before building, not during - included: Where does this data live? Who can access it? How long is it retained? What happens if a user enters identifying information about a real employee? Is there a data breach response process?

For HRWise.com.au, I made the deliberate design decision to keep the triage tool non-identifying - users describe situations, not people. That was partly a privacy decision and partly a product design decision: the tool is more useful when it is triage, not case management.

What this means for you: Before you build anything that touches HR data, read the Australian Privacy Principles. Understand what a notifiable data breach requires. And if you are building for other organisations, consider whether you need a Data Processing Agreement. Get legal advice on this specifically - it is inexpensive relative to the risk.

---

7. The hardest part is not building it - it is trusting yourself to launch it.

I delayed the first real launch of HRWise.com.au by six weeks. Not because it was not ready. Because I was not ready.

The specific anxiety was: what if an HR practitioner uses this and gets it wrong? What if the triage flags something incorrectly and a manager makes a bad decision?

What resolved it was being honest about the scope. HRWise.com.au is a triage tool, not a legal advice service. The disclaimer is clear. The purpose is to help HR practitioners and managers structure their thinking before they decide whether to escalate, investigate, or take action - not to replace the decision itself.

SHRM data from 2026 finds that 92% of CHROs are accelerating AI integration into HR functions - which means the question for most organisations is no longer whether to use AI, but how to use it responsibly. A tool that is clear about its purpose and its limits is more responsible than no tool at all.

The practitioners I have spoken with since launching have not been worried that the tool does too much. They are relieved that it does one thing clearly.

What this means for you: Scope your tool explicitly. Put the scope in the interface itself - "this tool helps you triage, not decide." Clarity about what a tool is for is not a weakness. It is the thing that makes it trustworthy.

---

8. This changed how I see my HR expertise - and yours should too.

Before I built HRWise.com.au, I thought of my expertise as something I delivered through consulting engagements, advisory work, and direct practitioner support.

Building the tool forced me to see it differently. My expertise is a system. It is a set of structured decisions, frameworks, questions, and judgement calls that I have refined over 14 years. That system can be documented. It can be encoded. And once it is encoded, it can be delivered at a scale and a price point that consulting time never could.

AI adoption in HR jumped from 26% to 43% between 2024 and 2025 according to SHRM - and teams using AI save approximately 20% of their work week, roughly one full workday, according to LinkedIn research. That is not efficiency for its own sake. That is capacity redirected toward the work that genuinely requires a human - the complex, the contested, the nuanced.

The HR practitioners who will thrive in the next five years are not the ones who resist these tools. They are the ones who understand that their expertise is valuable enough to encode, and who build the systems to do it.

You do not need three months and a redundancy payout to start. You need a problem you know better than anyone, a tool that can help you solve it, and the patience to learn through the confusion.

Start building.

---

---

Key Takeaways

1. You do not need to code. You need to be precise. The discipline of writing exact instructions for an AI agent will make you a better HR practitioner.
2. The tools have caught up. n8n, Supabase, Claude, Gemini - the ecosystem for non-developer builders is mature enough to build production-quality tools in 2026.
3. Your expertise is the product. The AI is the delivery mechanism. Encode your frameworks, your decision trees, your judgement. That is your IP.
4. Build the smallest useful thing first. One flow. One problem. One use case. Launch it. Then build what the feedback tells you to build.
5. The Australian ER/IR context is a competitive advantage. Global tools do not handle the Fair Work Act, modern awards, or psychosocial obligations well. Your local knowledge is the point of difference.
6. Privacy is not an afterthought. Read the Australian Privacy Principles before you build anything that touches HR data.
7. Scope your tool explicitly. Triage, not advice. A clear scope makes a tool trustworthy.
8. This is not about the tools. It is about recognising that your expertise is valuable enough to scale - and building the systems to do it.

---

Sources

1. [1] SHRM. State of AI in HR 2026. Survey of 1,908 HR professionals, December 2025. 39% of HR functions have adopted AI; 62% of organisations deploying AI somewhere in operations.
2. [2] CHRO Association and University of South Carolina's Darla Moore School of Business. 2026 CHRO Survey. 91% of CHROs rank AI and digitisation as their top concern. Cited in IMD: AI and the CHRO.
3. [3] LinkedIn. Global Talent Intelligence 2026. Teams using AI save approximately 20% of their work week (roughly one full workday). Cited in SQ Magazine AI Recruitment Statistics 2026.
4. [4] DronaHQ. Top 9 No-Code AI Agent Builders (2026). n8n added AI Agents as a native feature in late 2025. dronahq.com/best-no-code-ai-agent-builders/
5. [5] evalflow.com. 100+ HR Statistics for 2026. 92% of CHROs are accelerating AI integration; McKinsey found nearly 8 in 10 organisations have deployed AI tools.
6. [6] Office of the Australian Information Commissioner. Australian Privacy Principles. Privacy Act 1988 (Cth). oaic.gov.au/privacy/australian-privacy-principles/
7. [7] Founderreports.com. AI in the Workplace Statistics for 2026. 78% of organisations used AI in some capacity in 2025, up from 55% the year prior. Drawing on SHRM, Gallup, McKinsey data.