Most conversations about legal AI start in the wrong place.

They start with the technology: what the model can do, how fast it drafts, whether it hallucinates less than the last version, etc. Those questions are fine to ask. But they are not the questions that determine whether a tool actually changes how your team operates.

Newsweek’s General Counsel, Zahreen Ghaznavi, recently sat down to discuss her team’s experience evaluating and adopting AI for their in-house legal function. What struck me was not the list of features she mentioned. It was the decision architecture underneath: how she and her team moved from curiosity to trust, and from trust to genuine operational transformation.

That journey is the one most legal teams are navigating right now. And it reveals something

important about what this moment actually demands.

The Real Starting Point Is Not Technology. It Is Self-Diagnosis.

Ghaznavi’s team did not start by surveying every AI vendor on the market. They started by asking what the business needed from legal, and what gaps existed in how the team was currently delivering.

This distinction matters more than it sounds.

When you begin with the vendor landscape, you likely end up overwhelmed. Why? Because there are dozens of tools, each claiming to solve problems you may not even have. So, when you begin with your own operational reality (where the bottlenecks are, where the quality risks live, where your team spends time that does not match the value they should be delivering, etc.), you build a filter that makes the rest of the evaluation manageable.

Ghaznavi described this clearly: her team needed speed, legal-specific depth, and something they could use without becoming prompt engineers (most lawyers and legal ops barely have time to complete all of their work, let alone learn how to become a legal prompt engineer). This clarity by Ghaznavi allowed them to eliminate tools quickly and focus their energy on the ones that actually matched how legal work happens inside their organization.

But most legal teams skip this step. And that is where the frustration begins.

Trust Is Not a Feature. It Is a Process.

One of the most telling moments in Ghaznavi’s account was her description of vetting trust during the evaluation process.

Her team came into demos with questions they already knew the answers to. They watched whether the tool cited credible sources. They noted when a product hallucinated during a sales demo, and then they drew the obvious conclusion: if it fails when the vendor is trying to impress you, what happens when your team is relying on it at speed, under pressure, on a Thursday afternoon before quarter-end?

This is something I see consistently across legal teams evaluating AI. Trust does not come from a slide deck. It comes from putting the tool through conditions that mirror your actual work (messy documents, ambiguous questions, jurisdictions you are less familiar with) and seeing whether the output holds up.

Ghaznavi’s team sandboxed three tools before making a decision. They fed in real briefs, real discovery documents, real claim letters. They checked citations. They tested tone. They asked whether the AI could produce something that sounded like their team, not like a generic language model.

That level of rigor is what separates teams that adopt AI meaningfully from teams that buy a license and never use it.

The Transformation No One Talks About

The part of Ghaznavi’s story that resonated most with me was not the efficiency gains, though those were real. It was the shift in what her team was able to spend their time doing

Before adopting a legal AI platform, her team was spending significant hours on research, on processing unfamiliar areas of law, and on producing first drafts that consumed disproportionate time relative to their strategic value. With that time reclaimed, the team moved into a fundamentally different operating posture.

They became faster, yes. But more importantly, they became more strategic. They could scope issues before engaging outside counsel, saving money and sharpening the questions they brought to their external counsel. They could serve as true business partners, not because they worked harder, but because the infrastructure underneath their work had changed.

When Ghaznavi was asked what would happen if AI went away tomorrow, her answer was immediate: slower response times, more outside counsel spend, and less confidence making decisions across jurisdictions her team does not specialize in.

That is not a description of a nice-to-have tool. That is a description of infrastructure.

What This Means for Legal Teams Right Now

Ghaznavi’s experience illustrates a pattern I see across every legal team that successfully adopts AI. It is not about finding the most impressive technology. It is about finding the tool that fits the way your team actually works, and then building the trust and habits that turn a software license into operational capability.

Three things consistently separate the teams that get real value from the ones that do not.

First, they start with their own needs, not the vendor landscape. They know what problems they are solving before they see a single demo.

Second, they test rigorously. They sandbox. They bring real work into the evaluation, not hypothetical prompts. They ask for time (two or three weeks, not a rushed one-week trial) because legal teams are busy, and a meaningful evaluation requires meaningful use.

Third, they involve the whole team. Ghaznavi described a process where every team member

had a voice in the decision. That matters because adoption is a team behavior, not an individual one. If the people doing the work every day do not feel invested in the choice, the tool sits unused.

The legal teams that will define the next era of in-house practice are not the teams with the biggest budgets or the most sophisticated technical knowledge. They are the teams that approach this moment with clarity about what they need, rigor in how they evaluate, and the willingness to let their workflow evolve.

The technology is ready. The question is whether your evaluation process is.