At Axiom's Q3 Legal Leaders' Breakfast, I spent the morning with a group of in-house legal leaders talking through the question that seems to come up in every legal department planning conversation right now: How do we get real value from AI?
The discussion covered a lot of ground, and the group was open about what is working and what still is not. The view that shaped the rest of the conversation was that AI changes how legal services are designed and delivered. That makes it a much bigger undertaking than a technology purchase, and it explains why so many teams are finding that licenses alone don’t move the needle.
The teams seeing the best results share a few habits. They start with a specific business problem. They simplify the underlying workflow before automating it. They build controls into day-to-day practice. And perhaps most importantly, they invest in their people.
Adoption and Impact Are Two Different Things
The first point we landed on was that adoption does not equal impact. A legal team can hold a large number of AI tool licenses and run several isolated experiments without changing how it delivers services or seeing any return on investment.
Several delegates described exactly this situation. Their teams have plenty of licenses but habitual use remains limited. Readiness and confidence can also vary a great deal within the same team.
Delegates shared several ways they’re tackling this. Some are starting with volunteers and practical use cases. Others are sharing peer examples. Several are varying access by role and need so that tools go where they will actually be used. And there was broad agreement that change management is a leadership responsibility.
AI Exposes the Problems in Your Current Processes
A recurring theme was that AI tends to surface problems that were already there. Inconsistent intake, fragmented knowledge, unclear ownership, and non-standardized workflows limit a team's ability to scale AI far more than the technology itself does.
This showed up most clearly in conversations about fit. Delegates found that generic contract review and workflow tools are unreliable when they lack business-specific playbooks, standards, and knowledge.
The fix starts before any tool is switched on. Delegates are mapping the workflow first, defining what “good” looks like, and preparing approved templates and playbooks. They’re also investing time in prompts, calibration, and feedback loops, and making a deliberate effort to capture institutional knowledge. Contract lifecycle management is one example of how legal teams can put that process work into practice.
One strategy idea that came out of the morning captures this well: Design the service before selecting the tool. Identify the legal service or workflow you want to improve, then separate the work into what can be self-served, what can be standardized, what can be automated, what should be escalated, and what must be reserved for expert judgment.
It also helps to build an AI readiness baseline. That means assessing process consistency, knowledge quality, data access, ownership, risk appetite, skills, and how well any new tool will integrate with the systems people already use every day. Legal operations professionals can help teams assess and improve those workflows.
Start With a Defined Problem
Nobody in the room was looking for one platform to solve every legal need. Teams are prioritizing narrow, repeatable, high-volume tasks where a clear baseline and outcome can be measured.
The strongest near-term opportunities the group identified were reducing low-value work, improving intake and knowledge retrieval, and enabling carefully controlled self-service for the business. Specific use cases discussed included intake quality checks, knowledge retrieval, document summaries, first drafts, low-complexity contract work, and email triage.
A related idea is to make institutional knowledge reusable. Recurring advice, playbooks, templates, and past decisions can be converted into governed knowledge that lawyers can query safely. Where appropriate, the business can query it too.
The most effective transformations begin with a clear operational challenge.
Governance That People Will Actually Follow
Governance produced one of the more nuanced discussions of the morning. Delegates recognized a real tension here. Overly restrictive rules can push people toward unapproved tools, while uncontrolled use creates risks around confidentiality, errors, and weak auditability. I wrote about shadow AI earlier this year, and it was clear from this conversation that legal leaders are feeling that tension firsthand.
The teams handling it well are using clear risk tiers, approved tools, and data privacy rules. They are adding source checks, human sign-off, audit trails, and escalation points to their workflows, and they are reviewing those controls regularly as tools and use cases change.
Supervision works best when it is designed into the workflow from the start. That means specifying what a tool may do independently, what requires review, and who owns approval. It also means training lawyers to test, challenge, and improve AI outputs. A review step only protects the organization if the reviewer is genuinely engaged with what they are reviewing.
Human Judgment Remains Central
Throughout the morning, we repeatedly described AI as an assistant that accelerates work. Lawyers remain responsible for context, challenge, decisions, and accountable sign-off.
This matters most in complex work. Delegates noted that enterprise-wide tools can be less dependable in specialist, regulated, or multi-jurisdictional matters. The approach many are taking is to use AI for retrieval, organization, and first-pass analysis, and to reserve nuanced regulatory interpretation and final advice for appropriately experienced lawyers.
It is also worth remembering that a lawyer's ethical and professional obligations apply fully to AI-assisted work. Duties of competence, confidentiality, supervision, and candor don't change with the tools. The ABA addressed this in Formal Opinion 512, and many state bars and courts have issued their own guidance. Because requirements vary by jurisdiction and continue to evolve, legal leaders should build their AI governance around the rules that apply to their lawyers.
The Business Is Arriving With Its Own AI Answers
One of the most relatable topics of the morning was how the relationship between legal and the business is changing. Business colleagues increasingly come to legal with views they have already formed using AI tools.
Delegates described receiving lengthy AI-generated letters and opinions, some of them inaccurate, that take a lot of time to scrutinize. Legal teams now have a role in validating those inputs, correcting missing nuance, and helping colleagues use AI tools safely.
The practical responses here were straightforward. Set clear expectations for how the business uses AI on legal matters. Require clear facts and sources. And teach colleagues to share the prompt and underlying materials along with the output so legal can see how a conclusion was reached.
The broader lesson is that legal should help shape enterprise AI decisions and design self-service options together with business users so that tools are never implemented in isolation.
The Hardest Question: How Will Junior Lawyers Learn?
If there was one issue the group did not fully resolve, it was this one. Routine drafting and review have long been the training ground where junior lawyers develop attention to detail, legal judgment, and commercial instincts. As AI takes on more of that work, that pathway may weaken.
The approach delegates described is to redesign training so the learning survives even as the tasks change. Ideas raised included supervised review of AI output, involving junior lawyers in playbook maintenance, asking them to explain the rationale behind their work, rotations, and deliberate exposure to complex business conversations.
This connects to a wider workforce question. Legal teams will need an explicit plan for new skills in legal operations, knowledge engineering, process design, data stewardship, and AI-enabled legal supervision. Protecting development routes for junior talent belongs in that plan. Because this issue is still unsettled, it deserves deliberate attention now.
Measure Value in Business Terms
Faster drafting is only valuable if it leads to something the business cares about. The group identified the outcomes that count as better speed, extra capacity, stronger risk management, a better stakeholder experience, and avoided external spend.
Proving that value is harder than it sounds. Delegates pointed out that saved time is difficult to capture. There was also a real concern that time savings can become a rationale for future headcount pressure if they are not clearly linked to improved outcomes.
The teams getting ahead of this are setting baselines for volume, turnaround, rework, cost, and satisfaction before they deploy a tool. They then track workflow-level outcomes, quality, and external spend avoidance alongside usage figures. When reporting to leadership, useful measures include contract velocity, quality, rework, capacity redeployed, stakeholder experience, risk incidents, and external spend avoided. It also pays to explain clearly where released time is being redirected, so leadership sees the capacity being put to work.
Rethinking the Economics of Outside Counsel
There’s a widespread expectation that AI should reduce external legal spend. Delegates noted, however, a lack of transparency from law firms about their own AI efficiencies and how those are reflected in pricing.
The approaches shared here focused on open conversation. Legal leaders are discussing staffing and technology assumptions directly with their firms. They are identifying work that could be insourced or moved to alternative delivery models. And when negotiating, they are asking for value and outcome measures in addition to lower hours.
Rising law firm costs are pushing legal teams to rethink their approach to legal resourcing.
Questions to Ask Your Team
To close the session, we discussed the questions that can help a legal leader clarify their own AI strategy. They are a useful starting point for any team:
- Which legal workflows are sufficiently repeatable, high-volume, and low-risk to improve first?
- What will make the team and its knowledge genuinely AI-ready?
- Where should legal draw a hard boundary, and where should it enable controlled experimentation?
- How will the team develop judgment, creativity, and commercial understanding as routine work changes?
- Which business outcomes will demonstrate value credibly to senior leadership?
- What level of self-service will improve the experience for the business without creating unmanaged risk?
There’s no one-size-fits-all answer to AI in legal. The right pace and the right controls depend on each organization's risk profile, complexity, and culture.
The common message from the morning was clear, though. Redesign work around people and processes first. Then use AI to create more capacity for the things that matter most in legal: judgment, trust, and partnership with the business.
Frequently Asked Questions
What is the best way for a legal department to start using AI?
Start with a defined business problem. Legal leaders at Axiom's Q3 breakfast are prioritizing narrow, repeatable, high-volume tasks such as intake quality checks, knowledge retrieval, document summaries, first drafts, low-complexity contract work, and email triage, where a clear baseline and outcome can be measured.
Why do AI tools underperform in legal departments?
Common causes include low habitual use, inconsistent intake, fragmented knowledge, unclear ownership, and non-standardized workflows. Generic contract review and workflow tools are also unreliable without business-specific playbooks, standards, and knowledge.
How can legal teams govern AI without driving shadow use?
Use clear risk tiers, approved tools, and data rules, and build source checks, human sign-off, audit trails, and escalation points into workflows. Overly restrictive rules can push people toward unapproved tools, so controls should be reviewed regularly.
How should legal departments measure AI value?
Set baselines for volume, turnaround, rework, cost, and satisfaction, then track outcomes such as contract velocity, quality, capacity redeployed, stakeholder experience, risk incidents, and external spend avoided. Explain where released time is being redirected.
How will AI affect junior lawyer development?
Routine drafting and review have traditionally built junior lawyers' judgment and commercial instincts. Legal teams are redesigning training through supervised review of AI output, playbook maintenance, explaining rationale, rotations, and deliberate exposure to complex business conversations.