Why Trust Matters More Than Ever in AI: How We Built Phronesis Differently

Artificial Intelligence is transforming almost every profession. It can summarise documents, draft emails, analyse information and answer questions in seconds. For businesses, that promise is incredibly attractive.

But there is a problem.

When the advice you receive could affect someone's livelihood, your reputation, or even result in an employment tribunal, “probably right” simply isn't good enough.

At Phronesis, we didn't start by asking, “How can we build another AI chatbot?”

We started with a different question.

“How do we build an AI advisor that people can actually trust?”

That question has shaped every decision we've made.

The challenge with general AI

Most people have now used tools like ChatGPT, Gemini or Claude.

They're impressive. But they're also designed to answer almost anything. One minute they're explaining quantum physics. The next they're helping write a holiday itinerary. Then they're giving advice on employment law.

That versatility is their strength, but it can also be their weakness!

Large Language Models don't actually “know” employment law. They predict the next most likely words based on patterns learned from huge amounts of information.

Most of the time this works remarkably well. Sometimes it doesn't. And when it doesn't, the consequences can be serious.

The biggest problems with AI advice

When we began designing Phronesis, we identified several risks that worried us.

1. AI can sound confident when it is wrong

Perhaps the biggest problem with AI is that it often delivers incorrect information with complete confidence. Indeed people don't realise that AI has been built to be mostly positive about situations and issues it might be used for. This builds a sense of connectiveness and usefulness that isn't always reflective of the accuracy of the responses it provides.

To someone without HR expertise, it's almost impossible to tell whether an answer is entirely correct, partially correct or simply incorrect. In employment matters, acting on inaccurate advice can lead to:

• unfair dismissal claims
• discrimination claims
• breach of contract
• expensive tribunal proceedings
• damaged employee relations

We felt users deserved more transparency.

2. Every workplace is different

Employment law is only part of the picture. Every organisation has:

• different contracts
• different policies
• different procedures
• different collective agreements
• different ways of working

A generic AI doesn't know any of these things. Without recognising that limitation, advice can quickly become misleading.

3. HR decisions are rarely black and white

Many workplace issues aren't simply legal questions. They involve:

• judgement
• context
• evidence
• fairness
• competing perspectives

No AI should pretend it can determine the truth from one side of the story.

4. Some situations are simply too risky

Certain HR issues deserve specialist advice from experienced professionals. Examples include:

• discrimination
• whistleblowing
• TUPE
• collective redundancy
• settlement agreements
• employment tribunal claims

These are situations where organisations should slow down not accelerate to an outcome based on guidance provided by AI.

5. Users don't know what they don't know

Perhaps the most overlooked issue is this:

People often ask the wrong question.

Instead of asking:

“Can I dismiss this employee?”

The more important contextual questions might be:

“How long have they worked here?”

Or:

“Do they have a disability?”

Or:

“What does your disciplinary policy say?”

Without asking those additional questions, even a technically accurate answer may be inappropriate if the context for action isn't understood.

Building trust instead of simply building AI

Rather than trying to make AI appear more intelligent, we focused on making it more trustworthy. That required building safeguards around the underlying technology.

We call these our guardrails.

Rather than allowing the AI to answer everything without question, these guardrails are designed to improve reliability, transparency and user confidence. They are a core part of the Phronesis platform and reflect our governance-led approach to AI-supported HR advice.

So what makes Phronesis different?

We tell you how confident we are

Instead of presenting every answer as equally certain, Phronesis includes a confidence rating with every response.

• Sometimes the answer is clear.
• Sometimes assumptions have been made.
• Sometimes more information is needed.

Knowing the difference matters.

We explain where information comes from

Whenever possible, Phronesis references recognised sources such as:

• Employer policies
• UK legislation
• ACAS guidance
• relevant case law
• government guidance

Users shouldn't have to simply “trust the AI.” They should be able to understand why the advice has been given.

We never pretend uncertainty doesn't exist

Employment law isn't always straightforward.

• Sometimes case law conflicts.
• Sometimes guidance changes.
• Sometimes reasonable professionals disagree.

Rather than hiding uncertainty, Phronesis explains it. That honesty builds better decisions.

We distinguish fact from opinion

One of the easiest mistakes AI can make is blending:

• facts
• assumptions
• interpretation

Into one seamless answer.

Phronesis deliberately separates these wherever possible. That allows users to understand:

• what is known
• what has been assumed
• what is professional judgement

Those are very different things.

Trust is becoming the real differentiator

As AI becomes cheaper and more widely available, simply having an AI assistant won't be enough. Anyone can connect a chatbot to a language model.

The real question is:

Can you trust the answer?

That trust doesn't come from bigger models.

• It comes from thoughtful design.
• It comes from transparency.
• It comes from recognising uncertainty.
• It comes from knowing when to ask more questions.

And sometimes, it comes from saying:

“This needs a human expert.”

Looking ahead

Artificial intelligence will undoubtedly transform professional services over the coming decade.

But we believe the winners won't be those who produce the fastest answers. They'll be the organisations that produce the most trustworthy ones.

At Phronesis, that's exactly what we're trying to build. Not just an AI assistant. A responsible AI HR adviser.

Because when people decisions affect real lives, trust isn't an optional extra.

It's the product.