The human remains in command.
I am optimistic about artificial intelligence because I am serious about the work.
AI can help organize, compare, draft, test, and surface possibilities. It does not decide what a client should do. It does not determine which fact deserves belief. It does not own a deadline, make a representation to a tribunal, or accept responsibility when something goes wrong.
A person does.
AI can accelerate the work. It cannot own the judgment.
Treat AI like a junior team member.
The most useful analogy I have found is a very fast junior team member whose work must be supervised.
Give it a defined assignment. Explain the standard. Review what it produces. Challenge its assumptions. Send it back when the work is incomplete. Never confuse speed, confidence, or polished prose with competence.
The analogy has limits. A junior professional understands that a client is a person, develops judgment through experience, and bears real responsibilities. An AI system does none of those things. It can produce a fluent answer without understanding the answer—and without knowing when it is wrong.
A fluent answer is not a verified answer.
Where the value can be real.
Within an approved and secure workflow, AI may help a legal team organize large records, identify patterns worth investigating, build an initial chronology, compare drafts, generate lines of inquiry, test counterarguments, and surface potential avenues for further research.
The benefit I care about is not output for output's sake. It is the opportunity to spend more professional time on evidence, strategy, judgment, and communication with the client.
Used responsibly, these tools can create the conditions for deeper and more cost-efficient advocacy. That is a possibility, not a promise. The result depends on the tool, task, data, training, and quality of human review.
Training is part of the infrastructure.
Technology policy alone does not create judgment. Training does.
Southworth PC describes its internal Mindset training system as a curriculum of more than 400 modules. Its purpose is broader than teaching people how to operate software. It helps team members understand the work, the standards, and the reasoning well enough to recognize when an answer is incomplete, a source is unreliable, or an assumption does not fit the record.
A person who cannot evaluate the work is not supervising it.
Our research principle.
We have developed a proprietary research workflow intended to help lawyers find, compare, and test relevant authorities against a matter's record. I will not publish its mechanics, vendors, or security controls. The public commitment is what matters: an AI-generated research lead is not treated as law.
A lawyer must trace every material proposition to an authentic, current source and evaluate its jurisdiction, procedural posture, subsequent history, contrary authority, and relationship to the client's actual facts.
The tool may help locate a path. The lawyer must inspect the ground and confirm where it leads.
Confidentiality comes before convenience.
No one should assume that every AI tool handles information the same way. A legal team must consider terms, data retention, access controls, security, sharing, training practices, and the people or vendors that may receive information.
For federal employees and other members of the public, the safest default is straightforward: do not paste case details, personnel or medical records, legal advice, privileged communications, or other sensitive information into a public AI chatbot.
Disclosing information to a third-party system may create confidentiality or privilege consequences depending on the facts and governing law. Do not assume that deleting a chat, changing a setting, or removing a name resolves every risk.
The human-control checklist.
- Define the task.
Identify the purpose, stakes, limits, and whether AI is appropriate at all.
- Protect the information.
Use an approved tool and workflow, minimize sensitive data, and understand how the system handles it.
- Direct the work.
Give the system a bounded assignment and a clear source hierarchy. Do not ask it to make the professional decision.
- Challenge the output.
Look for missing facts, contrary arguments, bias, and unsupported assumptions.
- Verify independently.
Open and confirm every material source, citation, quotation, date, jurisdictional point, and statement of the record.
- Own the result.
A lawyer makes the decision, edits the work, approves any communication or filing, and remains accountable.
The standard is accountable progress.
Responsible adoption means understanding the technology, protecting information, defining human oversight, training the people using it, testing and monitoring the workflow, verifying the output, and preserving independent professional judgment.
The measure of responsible AI is not how much work we can automate. It is whether technology helps us serve people more thoughtfully while preserving the duties they came to a professional to receive.
The machine can assist. The human must understand, decide, and answer for the result.