Anthropic's Own Data Says Expertise Wins. Law Firms, Take Note.

The AI Company Just Published Proof That Experts Still Win

On June 16, 2026, Anthropic published a research paper titled Agentic coding and persistent returns to expertise, based on a privacy-preserving analysis of approximately 400,000 Claude Code sessions from approximately 235,000 people collected between October 2025 and April 2026. The headline finding is blunt. Novice users reached Anthropic's strictest measure of verified success only 15% of the time, while intermediate and expert users reached verified success between 28% and 33% of the time, according to coverage of the study.

Here is the part most coverage is missing. The company that profits when more people use AI just published evidence that AI roughly doubles the productivity of experts relative to beginners. This is not a story about technology replacing skilled professionals. It is a story about technology widening the gap between those who know what they are doing and those who do not.

For law firm leaders deciding how to deploy legal AI, that distinction carries real consequences. Here is what the data shows, why it matters for your firm, and what to do about it.

What the 400,000-Session Dataset Actually Found

Claude Code launched as a limited research preview on February 24, 2025, and became generally available on May 22, 2025. It reached $2.5 billion in annualized run-rate revenue, having more than doubled since the beginning of 2026, with enterprise use grown to represent over half of all Claude Code revenue. This is not a fringe research tool. It is one of the fastest-growing enterprise software products in history.

The usage data shifted in a revealing direction over six months.

• The share of sessions spent fixing broken code fell from 33% to 19%.
• The share spent operating software grew from 14% to 21%.
• Writing and data analysis roughly doubled, from about 10% to 20% of sessions.
• The estimated economic value of the average session rose by 27%.

Users stopped using the tool to clean up messes and started using it to do higher-value work. That shift did not happen because the model got smarter on its own. It happened because users got better at directing it.

The expertise effect held across professions. Software-related users reached verified success in about 30% of sessions, compared with 26% for users from other professions. The gap is real, but it is narrow enough to close with deliberate skill-building.

The Distinction Law Firms Keep Getting Wrong

Most firms treat AI adoption as a procurement decision. Buy the license, give it to associates, count the hours saved. Anthropic's data exposes why that approach underperforms.

Tooling is not skill

A Claude Code seat does not produce a 33% success rate. An expert using a Claude Code seat does. The verified-success gap between novices at 15% and experts at 28% to 33% is not a function of the software. It is a function of the human steering it. The same dynamic governs legal work. An associate who does not understand a fiduciary duty analysis will not produce a sound one by prompting an AI to write it.

The returns compound in the wrong direction for the unprepared

If AI doubles the output of skilled practitioners, the firms that win are the ones that pair the tool with deep subject-matter expertise and disciplined workflow design. Firms that buy the software and hope it substitutes for judgment will widen the gap in the wrong direction. They will produce more output of lower reliability, faster.

Verification is the whole game

Note that Anthropic measured *verified* success, not raw output. Code that runs is not code that is correct. A memo that reads well is not a memo that is right. The professionals who extract value are the ones who can verify the work product. In legal practice, that verification is the lawyer's core duty, and no model removes it.

What to Do Before Your Competitors Figure This Out

The market is already pricing in AI as a durable revenue engine. Anthropic raised $30 billion in a Series G round on February 12, 2026, at a $380 billion post-money valuation, and the company confidentially submitted a draft Form S-1 to the SEC on June 1, 2026 for a proposed initial public offering, though the number of shares and the offering price have not yet been set. The capital is committed. The question for service firms is execution.

First, invest in workflow design before tool selection. The session data shows value rose 27% as users learned to direct the tool toward higher-value tasks. Design the workflow that captures that shift. The license is the easy part.

Second, pair AI with expertise, not in place of it. The 15% versus 33% success gap is the entire argument against using AI to thin out senior judgment. Deploy it to amplify experienced lawyers, not to let juniors skip the learning.

Third, build verification into every AI-assisted deliverable. Anthropic measured verified success for a reason. Establish review protocols that confirm correctness, not just fluency, before any AI-assisted work product leaves the firm.

Fourth, train deliberately. The expertise premium is learnable. Software users beat other professions by only four points. A structured training program closes that gap faster than waiting for staff to figure it out alone.

Key Takeaways

• AI widens the expertise gap; it does not close it. Anthropic's analysis of approximately 400,000 sessions found novices succeed 15% of the time versus 28% to 33% for experts.
• Value follows skill, not seats. The estimated economic value of the average Claude Code session rose 27% as users learned to direct the tool toward higher-value work between October 2025 and April 2026.
• Verification is the differentiator. Anthropic measured verified success, not raw output, which mirrors the lawyer's non-delegable duty to confirm that AI-assisted work is correct.
• The capital thesis is settled; execution is not. With a $380 billion post-money valuation from its February 2026 Series G and a confidential SEC S-1 filed June 1, 2026, the market has bet on AI durability, leaving service firms to compete on how well they pair the tool with human judgment.

The Model We Are Building

The lesson from Anthropic's own data is one we have built our firm around. AI does not replace legal expertise. It rewards it. The firms that thrive will be the ones that treat AI as an amplifier of skilled judgment and build verification into every workflow.

This is the model we are building at FinTech Law: an AI-native securities practice where technology amplifies attorney expertise rather than substituting for it. If your firm or company is deploying AI into legal, compliance, or product workflows and wants to do it in a way that compounds expertise instead of diluting it, we would welcome the conversation. Contact us to schedule a consultation.

This blog post is for informational purposes only and does not constitute legal advice. No attorney-client relationship is formed by reading this content. If you need legal advice, please contact a qualified attorney.