Whether two artificial intelligence agents are better than one depends upon what you ask them to do. Attorneys use ChatGPT and Claude as individual litigation assistants, but using both artificial intelligence agents in the same case can be more effective than using just one of them.
I was already using ChatGPT in my defense of a lawsuit to read the pleadings, analyze the affirmative defenses and counterclaims, research the law, compare allegations with documents, characterize transactions, identify discovery opportunities, develop arguments and criticize those arguments.
Before using that analysis to develop litigation documents, I decided to prompt Claude, another AI agent, to independently analyze the same source materials. I could then compare the conclusions reached independently by the two agents and resolve any material differences. Only after that process would ChatGPT and I develop the litigation work product. The completed work product could then be submitted to Claude for another independent examination.
Lawyers have always understood the value of another set of eyes on a case. Why not use artificial intelligence agents the same way?
Putting Claude to Work
There are two different applications for a second artificial intelligence agent in litigation.
The initial independent analysis occurs before the analysis has been used to develop litigation documents. I gave Claude the same source materials I gave ChatGPT and used essentially the same prompts. Claude did not receive ChatGPT’s analysis or know what conclusions ChatGPT had reached. The two artificial intelligence agents independently examined the same materials.
After Claude completed its analysis, I compared its conclusions with those already developed using ChatGPT. Claude independently identified several of the same important issues, and it also identified matters that deserved additional consideration. Claude also made questionable assertions, but its independent conclusions provided a meaningful challenge to conclusions already reached using ChatGPT.
A material disagreement did not establish which artificial intelligence agent was correct. It identified an issue requiring further examination. After resolving the material differences between the two analyses, ChatGPT and I developed our litigation work product.
Agreement Is Not Verification
If two artificial intelligence systems agree with each other, that agreement does not establish that their common conclusion is correct. Disagreement may be even more useful because it identifies issues the attorney must examine more carefully. Agreement may provide additional confidence in a conclusion, but it does not verify it. Disagreement requires the attorney to determine why the agents reached different conclusions and what the record and governing law actually establish.
Independent Examination of Work Product
The second application begins after the initial analyses have been compared, any material differences have been resolved, and the attorney and the primary artificial intelligence agent have developed litigation work product such as a pleading, discovery demand or memorandum of law.
That work product is then added to the original source materials and provided to the second artificial intelligence agent for a new independent examination. The second agent does not receive the primary agent’s underlying reasoning. It examines the original source materials together with the completed work product and independently reaches its own conclusions.
The two applications are different. In the first, the two artificial intelligence agents independently analyze the same original source materials before substantial litigation work product is developed. Their conclusions are then compared and any material differences are resolved. In the second, the work product subsequently developed with the primary agent is added to the original source materials and submitted to the second agent for another independent examination.
The second agent can now consider what the attorney and the primary agent have done with the original materials. It may identify an inconsistency between the work product and the record or an unsupported factual proposition. It may identify something that was overlooked. It may reach a different legal conclusion. It may also be wrong.
I developed an Independent Examination Protocol defining the second agent’s assignment while preserving its independence. The protocol requires the second agent to distinguish facts, allegations, admissions, denials, inferences and assumptions; examine primary legal authority; identify burdens of proof; distinguish among procedural devices; search for adverse authority; distinguish substantive defects from stylistic preferences; identify uncertainty; and identify matters the attorney must independently verify.
The purpose of the protocol is not to tell the second agent what the primary agent concluded, but to ensure that the second examination is rigorous without influencing its conclusions. Any material disagreement identifies an issue requiring further verification by the attorney. The record establishes the facts. Primary legal authority establishes the law.
I use ChatGPT as my primary artificial intelligence agent and Claude as my second artificial intelligence agent. A lawyer who works principally with Claude can reverse the roles. The method does not depend upon which agent is used first. It depends upon preserving the independence of the two initial analyses and the independence of the second agent’s later examination of the work product.
ChatGPT does not supervise Claude. Claude does not supervise ChatGPT. Neither agent decides what the evidence proves. Neither determines what the law is. Neither decides what position the attorney should take. Neither signs a document and neither appears before the judge. The attorney does. Artificial intelligence gives attorneys another powerful way to do what they have always done. An artificial intelligence agent can examine thousands of pages, compare pleadings, search for inconsistencies, examine authorities, characterize transactions and challenge complicated legal analysis in a fraction of the time traditionally required.
But attorneys must still do what attorneys have always been required to do. Decide for themselves.
Are Two Artificial Intelligence Agents Better in Litigation?
September 2, 2026 | AI, AI in litigation
Whether two artificial intelligence agents are better than one depends upon what you ask them to do. Attorneys use ChatGPT and Claude as individual litigation assistants, but using both artificial intelligence agents in the same case can be more effective than using just one of them.
I was already using ChatGPT in my defense of a lawsuit to read the pleadings, analyze the affirmative defenses and counterclaims, research the law, compare allegations with documents, characterize transactions, identify discovery opportunities, develop arguments and criticize those arguments.
Before using that analysis to develop litigation documents, I decided to prompt Claude, another AI agent, to independently analyze the same source materials. I could then compare the conclusions reached independently by the two agents and resolve any material differences. Only after that process would ChatGPT and I develop the litigation work product. The completed work product could then be submitted to Claude for another independent examination.
Lawyers have always understood the value of another set of eyes on a case. Why not use artificial intelligence agents the same way?
Putting Claude to Work
There are two different applications for a second artificial intelligence agent in litigation.
The initial independent analysis occurs before the analysis has been used to develop litigation documents. I gave Claude the same source materials I gave ChatGPT and used essentially the same prompts. Claude did not receive ChatGPT’s analysis or know what conclusions ChatGPT had reached. The two artificial intelligence agents independently examined the same materials.
After Claude completed its analysis, I compared its conclusions with those already developed using ChatGPT. Claude independently identified several of the same important issues, and it also identified matters that deserved additional consideration. Claude also made questionable assertions, but its independent conclusions provided a meaningful challenge to conclusions already reached using ChatGPT.
A material disagreement did not establish which artificial intelligence agent was correct. It identified an issue requiring further examination. After resolving the material differences between the two analyses, ChatGPT and I developed our litigation work product.
Agreement Is Not Verification
If two artificial intelligence systems agree with each other, that agreement does not establish that their common conclusion is correct. Disagreement may be even more useful because it identifies issues the attorney must examine more carefully. Agreement may provide additional confidence in a conclusion, but it does not verify it. Disagreement requires the attorney to determine why the agents reached different conclusions and what the record and governing law actually establish.
Independent Examination of Work Product
The second application begins after the initial analyses have been compared, any material differences have been resolved, and the attorney and the primary artificial intelligence agent have developed litigation work product such as a pleading, discovery demand or memorandum of law.
That work product is then added to the original source materials and provided to the second artificial intelligence agent for a new independent examination. The second agent does not receive the primary agent’s underlying reasoning. It examines the original source materials together with the completed work product and independently reaches its own conclusions.
The two applications are different. In the first, the two artificial intelligence agents independently analyze the same original source materials before substantial litigation work product is developed. Their conclusions are then compared and any material differences are resolved. In the second, the work product subsequently developed with the primary agent is added to the original source materials and submitted to the second agent for another independent examination.
The second agent can now consider what the attorney and the primary agent have done with the original materials. It may identify an inconsistency between the work product and the record or an unsupported factual proposition. It may identify something that was overlooked. It may reach a different legal conclusion. It may also be wrong.
I developed an Independent Examination Protocol defining the second agent’s assignment while preserving its independence. The protocol requires the second agent to distinguish facts, allegations, admissions, denials, inferences and assumptions; examine primary legal authority; identify burdens of proof; distinguish among procedural devices; search for adverse authority; distinguish substantive defects from stylistic preferences; identify uncertainty; and identify matters the attorney must independently verify.
The purpose of the protocol is not to tell the second agent what the primary agent concluded, but to ensure that the second examination is rigorous without influencing its conclusions. Any material disagreement identifies an issue requiring further verification by the attorney. The record establishes the facts. Primary legal authority establishes the law.
I use ChatGPT as my primary artificial intelligence agent and Claude as my second artificial intelligence agent. A lawyer who works principally with Claude can reverse the roles. The method does not depend upon which agent is used first. It depends upon preserving the independence of the two initial analyses and the independence of the second agent’s later examination of the work product.
ChatGPT does not supervise Claude. Claude does not supervise ChatGPT. Neither agent decides what the evidence proves. Neither determines what the law is. Neither decides what position the attorney should take. Neither signs a document and neither appears before the judge. The attorney does. Artificial intelligence gives attorneys another powerful way to do what they have always done. An artificial intelligence agent can examine thousands of pages, compare pleadings, search for inconsistencies, examine authorities, characterize transactions and challenge complicated legal analysis in a fraction of the time traditionally required.
But attorneys must still do what attorneys have always been required to do. Decide for themselves.