Opus 4.6: AI agents are ready to replace lawyers in court

Remember when every practicing attorney laughed at AI agents' first attempts to draft a legal complaint? Those days are fading fast. This week's release of Opus 4.6 turned the agentic AI leaderboard upside down and forced the industry to rethink what's possible. While most users were debating image generation quality or chat response speed, the developers focused on the model's ability to hold extremely complex chains of reasoning. That is exactly what was missing for real legal work, where a single missing comma or a misread precedent citation in a multi-page contract can cost a company millions of dollars.

AI agents used to resemble talented but extremely careless interns. They could quickly find the right statute in a database, then instantly "forget" the context when they tried to apply it to a specific complex case. Opus 4.6 tackles this problem with a new attention architecture and improved planning. The agent no longer just reads text: it builds an internal dependency map that stays stable across hundreds of iterations. This lets the system carry a case from the initial analysis of a stack of documents all the way to a final defense strategy without the hallucinations that used to be the main obstacle to adopting neural networks in legal practice.

The legal field has always been seen as a bastion, shielded from automation by the complexity of human language and the nuances of law. But law is, at its core, a set of algorithms and rules applied to huge volumes of data. If an algorithm starts interpreting those rules better than the average law school graduate, the value of a human lawyer shifts from analysis toward ethics and personal representation. Opus 4.6 clearly shows that the technical side of the job — finding loopholes, checking compliance, and drafting hundreds of repetitive but legally precise letters — no longer requires expensive human involvement.

The market's reaction to this release is worth watching. Large law firms that ignored neural networks just a year ago are now rushing to roll out internal tools built on the new models. They understand that whoever first learns to delegate routine work to agents on the level of Opus 4.

6 will be able to cut costs by an order of magnitude or more. This is about more than savings; it changes the business model itself, turning hourly billing for "document review" into a relic. If a machine does in a second what takes a junior partner a week, what exactly is the client paying for?

The question is becoming rhetorical, and the answers will shape the future of the entire consulting industry.

Of course, a full-fledged "robot lawyer" in the courtroom is still a long way off because of legal restrictions, but the preparatory work is already almost entirely automated. Opus 4.6 shows striking accuracy on tasks that require multi-step planning. The model can test hypotheses on its own, query external databases, and adjust its actions based on the results. This is the real shift from a chatbot to an agent: one that acts autonomously to reach a set goal instead of just answering the user's questions.

We are entering an era in which competition between AI models moves from "who writes better poetry" to "who runs business processes better." Opus 4.6's lead on agentic benchmarks sends a strong signal to the entire industry. Where we once talked about AI assistance, it is now time to talk about replacing entire functional roles. Law is just the first swallow, and audit, logistics, and complex financial planning will inevitably follow. The pace of development shows that skepticism toward AI agents is becoming a luxury no business can afford.

The bottom line: Opus 4.6 has proven that agentic AI is no longer a toy for enthusiasts. Will lawyers be able to adapt to a reality where their main competitor doesn't sleep, doesn't ask for a raise, and doesn't misplace commas?