People or automation is a false choice

There is a false choice floating around legal AI: people or automation. Pick a side. Either you trust the software or you trust the humans. The firms getting genuinely ahead are refusing the choice and doing both, deliberately.

Start with a number that should give every “just buy the tool” pitch pause. McKinsey’s global research found that while roughly 65 percent of organizations now use generative AI regularly, only a small fraction have integrated it into more than a handful of functions, and only about one percent of leaders describe their AI rollouts as mature (McKinsey, 2024). Read that gap carefully. Adoption is everywhere. Mastery is rare. The tools are not the bottleneck. The operating model around them is.

Why does so much AI stall between purchase and payoff? Because AI does not run itself. Behind every system that actually delivers, there is unglamorous, continuous human work: prompt engineering, testing, monitoring, security oversight, quality review, and humans in the loop for the decisions that carry risk. That is the part most AI strategies quietly skip. A firm buys a capable tool, switches it on, and then discovers that nobody owns it, nobody is tuning it, and nobody is checking its output. Six months later it is shelfware, and the conclusion, wrongly, is that AI did not work.

The regulators have effectively codified the human-oversight requirement. The ABA’s Formal Opinion 512 makes clear that lawyers’ duties of supervision under Model Rules 5.1 and 5.3 extend to the use of AI tools, and that competence and oversight remain the lawyer’s responsibility, not the software’s (ABA, 2024). In other words, “the AI did it” is not a defense. Someone qualified has to be accountable for the work. That is not a constraint to resent. It is a design principle.

This is the thinking behind how we structure teams at V Group Inc, through what we call AI Pods and AI Pyramids. Rather than dropping a tool on a firm and walking away, we pair the right people with the right AI capability, sized to the actual work. A firm picks the team, sees the pricing transparently, and scales up or down as needs change. The model puts the human expertise and the AI capability together, on purpose, because that is what actually delivers.

A few things make this approach work.

It is sized to the work, not sold as one big package. Some firms need a lot of capability and a little support. Others need the reverse. A pod can be shaped to fit, and a pyramid lets you put senior judgment where it matters and leverage AI and junior capacity underneath it.

It keeps humans accountable for quality. The people in the pod are the ones reviewing output, refining prompts, and catching the issues a tool alone would miss. That is exactly the oversight the ethics rules require, built into the delivery model rather than left to chance.

It is flexible. Legal workloads are not flat. They spike with matters, deals, and litigation. A staffing model that flexes with demand is far more sensible than hiring permanently for a peak or leaving a tool unsupported in a trough.

And it transfers capability over time. Done well, this is not permanent dependency. The pod helps a firm build its own muscle, so the firm grows more self-sufficient as the engagement matures.

There is a talent angle worth naming too. With associate attrition running high and structural across the profession, leaning entirely on a thin internal bench to also operate AI is a fragile plan. A flexible, supported model gives a firm capability that does not evaporate the moment a key person leaves.

Here is the core belief underneath all of it. Capability, plus the people to operate it, is what makes AI stick. A tool on its own is a hopeful purchase. A tool with the right people around it is an operational advantage. The firms that understand this will quietly outperform the ones still waiting for software to run itself.

If your firm wants AI that is actually supported, not just installed, this is the model worth discussing. I will be at ILTACON 2026 in Nashville, at Booth #858. Come tell me where your last AI initiative stalled, and let us talk about what it would take to get it unstuck and keep it running.

Sources

  • McKinsey, The state of AI in early 2024 (about 65 percent use gen AI regularly; few have scaled it broadly or reached maturity): mckinsey.com
  • ABA Formal Opinion 512 (supervision duties under Model Rules 5.1 and 5.3), July 2024: americanbar.org
  • Gartner, Over 40 Percent of Agentic AI Projects Will Be Canceled by End of 2027 (operational support gaps as a cause of failure): gartner.com

#StaffAugmentation #LegalAI #LegalTech #ILTACON2026