Artificial intelligence is now commonplace. You’d be hard-pressed to go anywhere online and not see an AI assistant. Google, Amazon, and Meta (Facebook and Instagram’s parent company) are all pushing their AI tools on the masses. It’s impossible to do a simple search without the first result being a “Google overview.”

As AI becomes more and more normalized, it’s necessary to speak about how we reached this point and where we might be heading. If you follow the thread of AI, starting on November 30, 2022, when ChatGPT first went public, to the time of this article’s publication, you’ll see a rapid evolution from simple chatbots to advanced “agents” that can work together to solve difficult problems.

But what does this mean for lawyers who want to use AI in their practice? And what ethical implications must you consider before implementing AI tools in your work?

12 Months Ago: Chatbots Become Ordinary

About twelve months ago, chatbots were the primary option when it came to using AI. By their very nature, chatbots are excellent thinking partners. If you ask a question, they can return an answer. The chatbot format served as an excellent way to help with each step of a process. Chatbots have their shortcomings, however. This can particularly be seen when setting a larger goal or intention. Chatbots were not able to clearly develop and execute a plan, especially when working over a long period of time.

Six Months Ago: Agentic AI Is the Next Big Thing

Cut to six months ago. This is when the topic of agents and agentic AI started to take hold in Silicon Valley. With agentic AI, you can go beyond what a chatbot can do with the help of advanced AI models known as “agents.” By giving an agent a larger goal or intention, the agent can then plan out all the steps necessary to get from Point A to Point B. Then, the agent can execute those tasks and bring back a deliverable for the human user to review.

Three Months Ago: Claude Code & ChatGPT Codex Take Hold

Over the last three to four months, agentic AI has found its footing. With applications like Claude Code and ChatGPT’s Codex, agents have been able to translate what chatbots used to be able to do and extend those abilities into longer running tasks.

While agents are able to work harder and longer than basic chatbots, they still require some guidance in order to execute the tasks at hand without losing the original thread. This is where harness engineering comes into play.

Harness Engineering

Harness engineering can be seen as the “operating system” that runs the agents. Good harnesses typically follow this pattern:

• Orchestrator model: The current state-of-the-art LLM model (Claude Fable, ChatGPT 6.1 Sol) sits at the top. This is the model you will primarily interact with. You will describe your goal or intention to this model, and then the orchestrator model will create a detailed plan of all the steps to work through your desired process.
• The harness: Once you’ve approved the orchestrator model’s plan, then the harness will delegate the tasks to a fleet of sub-agents.
• Sub-agents: Sub-agents are typically lesser LLM models that are cheaper, faster, and more efficient to run than the state-of-the-art orchestrator models. Each sub-agent is only responsible for one specific task in the overall plan.

The orchestrator model also receives reports from the sub-agents and troubleshoots problems. Once the sub-agent’s work is complete, it is sent to the orchestrator model to synthesize the findings to deliver it to the individual for review.

The Advantage of a Good Harness

Harnesses give good guardrails and best practices so that agents are able to delegate these tasks efficiently. With a harness, tasks can run smoothly to deliver a good result. While this concept may sound complicated, implementing it is not.

Joshua Weaver, Director of Innovation at the State Bar of Texas, recommends this Github repository as an excellent place to start with orchestrator models and harnesses. It details "Claude Code skills that make Claude (Fable) the conductor of AI worker fleets."

If you have ChatGPT's Codex companion app and Claude Code, you can install orchestrators in one simple command. A google search for "add conductor to Claude Code" will steer you in the right direction. We recommend doing a bit more research on this topic to ensure you're following the right directions.

Putting Your Harness to Work

Agents and harnesses work best when they’re facing difficult, challenging problems, as they work for a long time in order to bring back a deliverable. If you have a large task that you’ve been putting off due to its complexity, consider using this harness to work through it. The results may surprise you!

A Word About Wrappers: Legal AI Tools vs ChatGPT and Claude

In general, legal AI tools use ChatGPT, Claude, or Google Gemini as their underlying large language model. Harvey, Clio, and MyCase all use Claude in this way. While these legal-branded tools do offer convenient, streamlined services, they are all “wrappers” for the main AI model. This means that it is possible to achieve the same benefits one may gain from a legal-branded AI tool by going directly through Claude or ChatGPT.

When using an AI tool like ChatGPT or Claude, instead of a legal AI tool, you will naturally face a steep learning curve but in time, you will be able to achieve the same or higher quality of work that you would if you were to use any legal-branded AI tool. It may not be as straightforward as a legal AI tool at the beginning, but by devoting some time to learn how the underlying AI model works, you can make it work for your exact needs, and you may even save a bit of money in the long run.

Always Use the Paid Pro Models When Using an AI Tool

If you’re already planning to pay for an AI tool, be it a legal tool or a more general LLM, it is highly recommended to subscribe to the pro model, instead of using the free options. Pro paid LLMs offer much more safety and customization options than the free models. A simple rule of thumb is that if you’re going to use an AI tool professionally, you should use the paid model.

The Ethical Implications Lawyers Face when Using AI

When deciding between using a legal AI tool like Clio or Darrow or going straight to ChatGPT or Claude, it is necessary to consider the ethical implications. Opinion 705 from the Professional Ethics Committee of the State Bar of Texas addresses the ethical implications that all lawyers must consider when using AI tools.

The State Bar of Texas AI Toolkit features an entire chapter dedicated to the ethics and professional responsibility standards that lawyers must contend with when using artificial intelligence in their law practice.

Opinion 705, which was issued by the Professional Ethics Committee for the State Bar of Texas in February 2025, addresses the ethical implications surrounding the use of generative artificial intelligence by attorneys. While the landscape of AI tools continues to evolve at a breakneck pace, attorneys must keep the following obligations top of mind when using AI tools, no matter how

Understand How Generative AI Functions

Outlined in Rule 1.01 of the Texas Disciplinary Rules of Professional Conduct, competence is central to the attorney’s obligations in respect to AI. In this context, competence requires the attorney to understand the function of generative AI in addition to possessing or acquiring the necessary skill to use AI tools effectively and ethically. Attorneys must not blindly rely on generative AI outputs. They must assess critically and verify the accuracy of all generated content.

Always Protect Client Confidentiality

Maintaining client confidentiality is a highly important concern attorneys must contend with when it comes to using AI. Since AI tools require detailed user inputs to function effectively, it’s necessary to exercise caution so as not to inadvertently disclose confidential client information.

There are risks associated with inputting sensitive client and case details into AI systems, as many of these systems inadvertently share information with third parties or use that information to train future AI models. To mitigate risk, it is necessary to properly vet generative AI tools for confidentiality safeguards. Train staff to ensure compliance with confidentiality rules as well.

Don’t Rely on AI Without Properly Verifying the Results

Oversight and supervision are some of the most crucial ethical considerations to keep in mind when using AI. Verifying anything generated by AI is a must, especially when it comes to legal work. As a lawyer, you must independently verify any and all information generated by AI. Doing so ensures accuracy and reliability. Opinion 705 states that using AI-generated content without proper verification could put attorneys at risk of violating rules related to fairness, honesty, and candor to the court.

Be Aware of Fair Billing Practices

The efficiencies gained through AI must benefit the client financially when using hourly billing methods. On one hand, attorneys cannot bill clients for hours not genuinely worked, even if the task has become streamlined through AI. On the other hand, however, reasonable costs associated with the direct usage of AI services and tools, such as subscription or usage fees, may be passed onto the clients with appropriate prior agreement.

As Legal AI Continues to Evolve, We Will Continue to Provide Guidance

When it comes to AI, it may seem as if there’s always a new issue or topic to delve into. If you feel this way, then you’re not alone. That’s why Texas Bar Practice and the Law Practice Management Department continuously produce content to answer the questions attorneys face when using AI.

We’ve already mentioned the State Bar of Texas AI toolkit, but the Law Practice Management department also produces various AI-related videos, including the twice-annual “Ask A Human” live webinar. In these live webinars, LPM attorney Trish McAllister speaks with lawyers and AI experts to answer AI-related questions.

In addition to the main “Ask a Human” webinars, we also publish monthly Q&A videos titled “Ask A Human: Answers to Unanswered Questions” where questions relating to AI ethics, confidentiality, and privacy are answered. Make sure to subscribe to our YouTube channel to stay up to date on future releases!

This blog post was based on this great video from attorney Joshua Weaver, Director of Innovation at the State Bar of Texas. Watch the video below: