AI is training our upper body, but who’s training our legs?
Most people who have spent time in a gym have seen someone who trains only their upper body. They look strong, and in many ways they are, until the moment they need balance, endurance, or stability.
Something similar may be happening in legal practice. As legal work speeds up, AI is increasingly being used for heavy procedural tasks like contract review, research, and first drafts. Indeed, among students, there’s a growing tendency to look for the strongest and fastest tools. These tools build the visible side of legal work: polished writing, clean structure, and quick answers. However, what they don’t build are the foundations underneath: judgement, nuance, market instinct, and critical thinking.
In gym terms, students risk training the upper body while skipping leg day.
The concern behind this article isn’t that AI will replace junior lawyers, it’s that AI may replace junior training. This point came through strongly in conversations with four lawyers including a partner at a mid‑tier firm and three junior lawyers at two Big Six firms. While their experiences were different, a few themes kept returning.
Thinking first, then bringing in AI
A consistent theme across our conversations was the importance of thinking first, and bringing in AI only after forming a view. The partner’s advice was clear: for now, juniors should still research the traditional way by running the searches, reading the cases, drafting the first version, and checking in with a supervisor. AI should only come in once the junior is satisfied with their own work, as a tool “to critique and to challenge [their] thinking”. The partner noted that using AI for a first draft can be helpful for people outside the law, but risky for juniors because the output can contain inaccuracies.
A junior lawyer at one Big Six firm described a similar routine. She completes the research or advice task herself, then uses AI to see whether it raises any points she missed. She prefers this order because seeing AI’s output first could bias her thinking.
Junior lawyers at another Big Six firm raised a related concern: if doing the work is how juniors learn, what happens when they no longer apply the knowledge themselves?
In this way, we found it interesting that both a partner and junior lawyers raised versions of the same worry, from opposite ends of the seniority ladder.
The problem you can't see
Another recurring point was that the risks people focus on are not always the ones that matter most. The partner noted that hallucinations are well known and can be reduced with trusted sources, but a bigger issue is omission: AI leaving out relevant information. Juniors can check that citations are real, but they are less equipped to notice what the AI did not cover, because, as the partner put it, “you haven’t thought around the problem yourself and you have not yet developed the necessary judgment or expertise to discern what might be missing”.
The partner explained that AI can be useful for senior lawyers working within their own area of expertise, because years of practice make them well placed to verify the output. Outside that area, the picture is different. One example was a corporate lawyer pasting a lease into an AI tool, skimming the comments, and moving on. The AI may not appreciate market or legal nuance, so the advice could be wrong. For juniors who have not yet built significant expertise, the partner described using AI as “particularly problematic”.
What the tools do well, and where they stop
The lawyers also gave concrete examples of where the tools help with routine work. The partner ran a prompt over a long, complex policy document and received a list of inconsistencies with suggested drafting fixes, many of them valid, which would have taken hours to identify manually.
Additionally, two juniors said that AI is genuinely helpful for procedural tasks that need consistency, and for pulling key terms out of PDFs in M&A reviews. In due diligence, they mainly use it to get a sense of direction. However, they pointed to clear limitations: the tools sometimes miss relevant issues or flag too much, and often struggle with nuance and deeper analysis.
They noted that the tools do not have a senior lawyer’s market practice or experience, and that one platform they use falls short on legal research, in their view not matching the skills of junior lawyers. Prompting also takes time, often up to 30 to 40 minutes, and the AI’s output still needs to be reviewed carefully.
The value of doing it the hard way
A junior lawyer said she was grateful she learned menial tasks the traditional way, such as trawling through documents to build a chronology. Those tasks built the analogue skills she now relies on to correct and check AI’s output. She cautioned students who default to AI without understanding the foundations of the work. The result can be less accurate and may need redoing, which cancels out any efficiency gain. Once those skills are in place, she said, the benefits are significant because juniors can handle time‑sensitive tasks for senior lawyers more efficiently.
She also noted the risk when AI checks AI and nothing original comes from the human author. The work can be incorrect, and it can sound flat or inhuman, which people can spot easily.
Engaging with AI, thoughtfully
The aim is to engage with AI carefully and deliberately. The partner described AI as an important development and said firms want juniors who understand how the tools work and who can recognise their risks, such as hallucinations, omissions, and confidentiality loss on public models. He emphasised that juniors should also understand the extent to which AI should be used, and when human judgement needs to take over. Junior lawyers at the Big Six firm described a similar push, with partners encouraging juniors to build capability so the firm does not fall behind.
Three ways to train your legs
1. Train the slow muscles first. Build the foundations first. Do the manual work before you automate it so you understand what the tool is doing, what it might miss, and what you need to check.
2. Use AI as a spotter, not a substitute. Draft first, then use AI to help you review. You can ask it to pick up inconsistencies, repetitions, conflicts, or points you didn’t raise. Prompting is a skill too, and delegating too soon can produce gaps that come from not thinking the problem through first.
3. Build judgement deliberately. When you propose an AI workflow to a supervisor, you should be able to explain the benefits, the risks, and the checks you will run inside and outside the tool. Think carefully about where critical thinking belongs in the process, and place it before the AI.
AI can make us feel senior before we have earned senior instincts. It builds the visible parts of competence and can hide the missing parts, and the gap may only show when we are the youngest in the room with an hour to deliver something.
None of the lawyers we spoke to argued for staying away from AI. They described a sequence: think first, bring in AI, then check its work, with those checks resting on foundations built the slow way.
In the end, the person who stands steady won’t be the one who looks strongest on top. It’ll be the one who didn’t skip leg day.