Nejc Novak, Co-founder of Arna, an AI legal agent for in-house teams, is a former attorney who began his career on complex restructuring cases arising from the financial crisis at a large law firm. In 2016, he founded his own boutique, which grew to operate in four countries, and he built it around supporting early-stage technology companies. That experience, and a first legal tech venture that did not work out, led him to build Arna. In this opening session of a series, he combined his personal story with a view on how AI will change who does legal work and how it is delivered.
Overview
Novak’s first company grew from a gap he saw in the market. Founders with no revenue and little cash tolerate most legal risks, but a few matter greatly, such as a corporate structure and shareholder agreement that venture investors will accept. In his experience, large firms in Europe struggled to serve this market, partly because the clients could not pay enterprise fees. His answer was to charge those with little, little, and those with more, more. Some early clients received free work, and most later became profitable relationships.
A health crisis shaped his thinking about risk. A stroke, caused by a spontaneous artery dissection in his neck, struck despite regular check-ups, fitness, and no identified risk factors. Doctors suspect a sports injury but are not certain. He argued that this challenged a transactional lawyer’s instinct that risk can be found early and managed. Some risks, personal or commercial, cannot be forecast or prevented.
On AI, Novak said he saw the potential of early language models when colleagues dismissed them as useless. He was clear that these are his views, not facts. He believes the profession will have failed if AI only lets large firms profit from needing fewer paralegals and juniors while clients’ experience stays unchanged. His interest is in-house teams, which have taken on more work for a decade. He does not expect that to mean less outsourcing; instead, he predicts a higher benchmark for external firms in both client experience and complexity. He also sees agents analysing large amounts of company information to tailor legal deliverables, though he does not see that happening yet.
His first customers were an early lesson. One, an employer of record provider, faced rule-driven administrative legal work that was hard to deliver quickly and accurately through people alone. Novak said a general AI assistant can cope at small scale but not at volume, which is where an agent is needed. He added that the learning curve in production is long.
Key Takeaways
- Match advice to what the client actually needs. Early stage founders accept many risks but care about a few, so identify those and be pragmatic elsewhere.
- Recognise risks that cannot be controlled. Novak’s experience suggests that lawyers who rely on early detection should also plan for what cannot be forecast, in their own lives and in their clients’ businesses.
- Choose a deliberate response to AI. He sees lawyers building tools, becoming expert users of AI assistants or stepping away, and he argues that drifting is not a strategy.
- Judge AI by client experience, not only internal efficiency. If only the production of deliverables changes, he says, the profession has missed the opportunity.
- Expect higher standards for external counsel. If in-house teams cover more work with AI, outside firms will be asked for more complex work and better service.
- Automate rule-driven work first. Administrative, repeatable legal tasks are where agents may free people for more creative work, but real learning only comes once a tool runs in production.