The question ‘to use or not to use’ artificial intelligence (AI) has run its course. For most practitioners, that debate is over. The highly publicised endorsements by the large law firms on their adoption of this technology has been an important industry-led signal. But caution remains as the hallucination sagas, both locally and abroad, are hard to forget. Now the question, ‘Should I use AI?’ has been replaced with ‘How do I use it responsibly?’.
This article will not be about prompting techniques, defining an LLM or the difference between foundation models. I want to present a framework for legal practitioners to amplify their skills with AI while mitigating the risks. Used correctly, the approach I present here should equip you to do just that.
The framework is simple: You are the pilot – always stay alert – especially during take-off and landing. A pilot generally requires the highest level of hands-on control and judgment during the ascent and descent of a flight. Once at cruise, they must remain alert but can rely more on ‘autopilot’.
For the legal workflow, the ascent is the starting phase of any matter. You must ensure a precise and detailed understanding of the facts, the contestation and the legal question at the heart of it. Do not delegate framing, judgment or the formation of legal opinions to a model. This is where the practitioner’s knowledge and experience should shine as they define the problem, mull over the potential paths to defend their client and attempt to find the gaps or weaknesses in a series of events.
Once in cruise, AI can help you with tasks that typically require heavy lifting or could benefit from an ‘outside perspective’. Use it to connect the threads between affidavits, stress test potential lines of argument, or run deep research. Your grasp of the brief must be strong enough to spot when an output is misconstrued or blatantly incorrect. And as you would with any intelligent junior colleague, always double check their work.
Finally, any output that leaves your desks must be reviewed by you alone and deemed ‘safe for landing’.
The difficulty with this framework is ensuring that you adopt it every time you use AI. Even as you gain confidence in the reliability of outputs, you should be able to defend, explain or verify every element of a final deliverable. Of course, an alternative approach is to only use it to help with low-level tasks like grammar, summaries or tone correction. This minimises the downside but also limits the upside. I want to share this framework to show all knowledge workers, but especially lawyers who face real reputational risk, that if you remain the pilot, it is possible to maximise the upside.
As AI evolves at pace, knowledge workers who want to push their own skills and abilities have good reason to be excited.
Julita Ramsunder Masters in Economics (Wits) is the Chief Operating Officer and cofounder at Adjunct in Johannesburg.
This article was first published in De Rebus in 2025 (Dec) DR 18.
The post Past the hype: Practical AI for people who think for a living appeared first on De Rebus.