Webinar recap featuring Dr. Matthias Artzt, Senior Legal Counsel at Deutsche Bank, hosted by Dylan Viljoen, Head of Delivery Growth at Radiant Law.
As AI adoption accelerates across legal functions, many organisations have moved beyond experimentation and pilot programmes. Legal teams now need to scale AI while keeping governance, cost and accountability firmly under control and the emergence of agentic AI has introduced a new layer of complexity.
Unlike traditional AI assistants that provide information or recommendations, AI agents can increasingly make decisions, interact with systems, trigger actions and operate with varying degrees of autonomy.
As legal and compliance teams evaluate where these technologies fit within their operating models, several critical questions are emerging:
These were some of the themes explored in Radiant Law's recent webinar. Drawing on his experience advising on AI governance, legal risk and emerging regulatory frameworks, Matthias shared practical insights into how organisations can govern AI systems responsibly while enabling innovation.
Here are the key takeaways….
One of the session's most important distinctions was the difference between AI assistants and AI agents.
Matthias proposed a simple three-part test:
Traditional AI tools primarily provide information. AI agents act on information.
This distinction matters because governance requirements change significantly once systems move beyond generating outputs and begin taking actions. Agents may trigger workflows, interact with business systems, instruct other agents or execute tasks with limited human intervention. As Matthias explained, AI getting an answer wrong is only the start of the problem. The bigger worry is what happens when an autonomous system acts on that wrong answer.
When one agent's output becomes another agent's input, errors can compound quickly. Governance therefore becomes less about managing individual model outputs and more about managing entire agentic ecosystems.
While much of the market discussion focuses on automation, Matthias highlighted three practical areas where AI agents are already demonstrating value.
Legal teams are increasingly using AI to:
These use cases remain some of the most mature and accessible applications for legal teams.
Agentic AI becomes particularly powerful when managing end-to-end processes. An agent can:
At Deutsche Bank, Matthias discussed how future agentic solutions could help manage data subject requests by identifying and extracting relevant personal data across multiple systems and repositories.
Regulatory requirements, contracts, policies and risks are not static. AI agents can (for example) continuously monitor the following and alert organisations when action is required:
This ability to monitor at scale may prove one of the most valuable long-term applications of agentic AI within legal and compliance functions.
A recurring theme throughout the discussion was the importance of evaluating business value before introducing AI.
Matthias cautioned against deploying agentic solutions simply because the technology exists. Instead, organisations should start by asking:
A task that takes five minutes once per month may not justify a sophisticated autonomous solution. However, a process performed thousands of times across multiple systems could present a compelling business case.
The lesson is simple: identify the business problem first and then determine whether AI is the appropriate solution.
One of the webinar's most thought-provoking insights challenged a common assumption: that greater autonomy always delivers greater value.
According to Matthias, this is often one of the biggest misconceptions surrounding AI implementation. For example:
As autonomy increases, so does the need for:
Autonomy has a price. Past a certain point, the cost of managing the risk outweighs what the extra autonomy delivers. The aim is the sweet spot where value, risk exposure and governance cost balance out.
Just as organisations should avoid over-automating, they should also avoid over-governing.
Matthias argued that governance frameworks should match risk. A low-risk internal research tool should not require the same controls as an agent capable of:
Governance costs money too. If a framework costs more than the risk it manages, nobody will stick with it.
The same goes for human oversight. Putting a person in the loop on everything wastes their time, so match the level of involvement to the risk. Higher-risk decisions get more human attention, and lower-risk tasks can run with more autonomy.
One of the most practical parts of the discussion focused on guardrails that organisations should implement before deploying agentic systems.
Matthias outlined six governance layers:
Clearly define:
Establish clear rules governing:
Define escalation thresholds requiring human review and intervention.
Implement safeguards covering:
Organisations should record:
Link every agent to an identifiable human owner or organisational function responsible for its operation. Importantly, build these controls into the system from the start. As Matthias summarised:
"Know your agent by design."
Embed guardrails in the architecture rather than adding them later as documentation.
A particularly interesting area discussed during the webinar was the role of evidencing governance decisions.
Drawing on themes explored in his article, Agentic AI Governance: Ten Reasons Legacy Controls Are No Longer Sufficient, Matthias argued that traditional governance frameworks were largely designed for passive software systems rather than autonomous actors.
As organisations delegate authority to AI agents, demonstrating that they've established appropriate controls may become increasingly important.
He highlighted blockchain as one potential mechanism for evidencing governance decisions and controls through immutable records, including:
While blockchain will not be necessary for every use case, the broader message was clear: governance must be demonstrable, not simply documented.
The discussion also looked ahead to emerging regulatory developments.
Matthias highlighted the EU Product Liability Directive (2024/2853), which entered into force on December 8, 2024 and must be transposed into member state law by December 9, 2026. The directive extends liability concepts to software and AI systems, creating important implications for organisations developing and deploying agentic technologies.
Courts have yet to test liability for AI-driven outcomes. However, legal teams should expect increasing scrutiny of:
This theme aligns closely with Matthias' article, Who Is Liable When Agentic AI Causes Loss?, which explores how responsibility may be allocated when autonomous systems contribute to harmful outcomes.
As agentic AI becomes more widely adopted, organisations will need to pay increasing attention to questions of accountability, responsibility and liability.
Perhaps the most forward-looking discussion centred on how legal professionals themselves may need to adapt.
In Matthias's opinion, lawyers will increasingly spend less time performing routine legal analysis and more time:
The legal professionals best positioned to succeed in this environment will be those who can operate across multiple disciplines, including:
The future lawyer may look less like a traditional legal specialist and more like a governance architect for increasingly autonomous systems.
One of the biggest lessons from the session was about where autonomy helps most. Good AI implementation means knowing where it creates real value, where you need proper controls, and how to balance innovation with accountability.
The organisations that get this right will be the ones that understand:
As AI settles into legal and compliance work, these questions stop being theory and become day-to-day priorities. Legal leaders will need to govern it well as autonomy, complexity and business reliance grow.
As legal teams move beyond experimentation and begin deploying AI at scale, governance, operating models and legal service design become increasingly important.
At Radiant Law, we help legal teams identify where AI can create value, structure governance frameworks, and integrate AI into legal operations in a practical, sustainable way.
Take the AssessmentYou might be wondering what your next steps should be. Let us guide you with three easy options:
We use cookies to monitor the performance of our website and to improve user experience. By continuing to browse our site, you agree to our use of cookies.
