Agentic AI: opportunities and challenges – and how to navigate them responsibly

Agentic AI represents a significant leap beyond traditional automation. These ‘digital coworkers’ are proactive, goal-oriented assistants – and the technology has profound implications across industry, with the potential to transform services by working in real-time, 24/7. The adoption of agentic AI also introduces a new set of challenges – with rapid evolution of the technology outpaces regulatory frameworks, careful consideration is needed of ethical, legal and operational risks. the emerging opportunities, challenges and approaches to best practice in this space?

In the past year or so, agentic AI has gone from a niche concept to one of the most talked-about frontiers in automation. It’s not just another chatbot trend. It’s a shift in how we think about software, automation and even the role of ‘digital workforces’ inside organisations.

IBM succinctly defines AI agents as “AI systems that are designed to autonomously make decisions and act, with the ability to pursue complex goals with limited supervision.”

If you’ve only ever used AI tools that respond to your prompts, think of agentic AI as the next evolutionary step – autonomous intelligence that can figure things out and take action.

Instead of waiting for instructions, agentic AI systems are proactive, goal-driven assistants. They can set objectives, plan steps, act and adapt to changes in real time – often with minimal human intervention.

This leap in capability opens incredible possibilities across industries. But it also raises new operational, ethical and governance challenges that organisations can’t afford to ignore.

From automation to autonomy

Traditional automation has always been about following rules: If A happens, do B.

Agentic AI flips that on its head. You can give an agent a broad goal – for example, “Reduce customer support ticket resolution time and improve satisfaction” – and it will figure out the intermediate steps, find the data it needs, take relevant actions and escalate to a human when necessary.

Imagine a customer support scenario where the AI:

Reads new support tickets
Pulls relevant data from your systems
Identifies the root cause of an issue
Corrects it where possible
Updates the customer on progress
Collects feedback
Escalates only when human expertise is required.

All of that, without a single line of code from your team.

That’s the power – and the challenge – of Agentic AI. You’re no longer just managing scripts or bots: you’re supervising a 24/7, never-tiring, highly capable digital coworker.

Where agentic AI is already making an impact

While the technology is still emerging, some industries are already seeing transformative results:

Government – autonomous agents can help citizens navigate complex services, engage proactively, answer questions, and pre-fill forms – making public services faster and more accessible, across multiple channels simultaneously
Healthcare – virtual assistants can schedule appointments, remind patients about medications, triage symptoms and route cases to doctors – around the clock
Financial services – lenders and insurers are using agents to automate credit underwriting and claims processing, adjusting risk models in near real time as new data arrives.

The common thread? These agents don’t just automate a single step – they manage entire workflows, adapt to new information and work alongside humans.

The scaling challenge

The hype around Agentic AI is justified – but so is caution. Unlike traditional automation, agentic workflows are non-deterministic. That means outcomes can vary each time, even with the same input. As you scale up, this introduces unique risks:

Accountability – who’s responsible if an agent makes a bad call that causes financial loss or harm?
Data security – are you inadvertently sharing sensitive or regulated data with public models?
Observability – how do you know why the AI made a certain decision?
Misuse – could bad actors exploit your agents for malicious purposes?
Bias and fairness – are outputs consistent and equitable across different user groups?

These aren’t hypothetical concerns – they’re already emerging in the first rollouts of the technology

A people-centric, responsible approach

If Agentic AI is the next leap in workplace automation, the way we adopt it matters as much as the technology itself. Introducing agentic co-workers into your organisation means new processes, evolved job roles, updated technology stack and different ways of serving customers. This is a design and change management challenge and approaching it as such will reap rewards in efficiency, adoption and capability development.

A practical framework to approach adoption

1. Start small, scale responsibly

Select a contained, non-critical use case with clear success measures. Run pilots with human oversight before handing over more autonomy. Learn and adapt.

2. Keep humans in the loop

Even when the AI can act independently, build escalation protocols for high-stakes decisions. In other words: automation should never mean abdication.

3. Plan for governance early

Establish cross-functional AI governance committees, with representation from compliance, legal, operations and ethics. Define policies for:

Acceptable use cases
Data handling and privacy
Escalation and incident response.

4. Design with people in mind

Map the user journey, including employees, customers and partners. Factor in accessibility standards and cultural differences. Use frameworks that make AI reasoning visible so users can trust decisions.

5. Test like it matters

Use diverse, structured testing with users of different profiles, roles and needs. Include scenario-based ‘red-teaming’ to stress-test against failures, biases, or adversarial inputs.

6. Monitor and adapt

Deploy monitoring systems to track AI decisions and performance over time. Hold regular stakeholder reviews and be prepared to pause or roll back functionality if risks emerge.

One creative tool we’ve explored is the ‘AI-sona’ – a twist on the design persona concept, used to explain the role and capabilities of AI agents to non-technical stakeholders. This simple reframing can prevent over- or under-estimating what the AI can do and helps teams think about agents as part of the workforce.

Technology choices that matter

Agentic AI isn’t just ‘plug and play’.. Your platform and model choices will affect performance, cost, and risk. Consider:

Using the right language model for the job – large general-purpose models are flexible but may underperform on domain-specific tasks. Specialised models can improve accuracy and reduce costs
Assuring data quality and integration maturity – bad or incomplete data will sink your agents faster than bad algorithms
Layering in guardian agents or meta-agents – these monitor for bad behaviour, enforce compliance and refine workflows as they run
Adopting zero-trust and encryption – security should be part of the build, not an afterthought.

The AI maturity curve: know where you stand

Not every organisation is ready to hand over mission-critical decisions to AI. Think of adoption as a maturity curve:

Assisted AI – low-stakes recommendations, full human control.
Augmented AI – I takes some decisions in well-defined areas, still with oversight.
Autonomous AI – AI manages high-stakes, complex workflows – with strict governance and limited (or no) human oversight.

The higher you go, the more value you can unlock – but also the heavier the compliance and risk-management burden.

The road ahead

The next big leap in agentic AI will likely come from self-learning models that can improve their own algorithms over time. We’re already seeing hints of this, with systems like Google’s AlphaEvolve optimising itself over thousands of generations.

For organisations, the challenge is not just what’s possible – it’s what’s responsible.

Technology will keep accelerating. Regulations, ethical frameworks and standards are racing to catch up, but they’re not there yet.

That means the burden is on today’s leaders to:

Experiment bravely
Plan deliberately
Keep people at the centre
Build safeguards in from the start.

The bottom line

Those who learn to blend its autonomy with human judgement will be the ones who turn it from hype into lasting business value.

If you’d like to explore how AI-sonas, governance frameworks or pilot programmes can help your organisation adopt agentic AI responsibly, we’d love to connect and share ideas.

Email clientservices@opencastsoftware.com and we’ll come back to you.

This post is based on a presentation by Gordon and Marianne delivered at the TechNExt 2025 festival in Newcastle on 18 June 2025.