The Displacement Is Already Happening. The Question Is Whether Anyone Will Say So Out Loud.

Let's start with the evidentiary honesty this story demands. The 90%+ recall rates from Harvey, CoCounsel, and Spellbook, and the 30-70% time reductions reported by in-house teams, are vendor-reported benchmark figures and client efficiency metrics from controlled or semi-controlled deployments. They are compelling and directionally meaningful. But they are not independent production deployment evidence measured across heterogeneous document sets, privilege determinations, or genuinely adversarial discovery contexts — the conditions where the evaluation-to-production gap matters most. We classify this as proximate evidence: it demonstrates capability at a level that makes first-pass document review automation technically viable, but it doesn't prove that firms are doing it at scale without attorney checkpoints, or that they'd say so publicly if they were.

The resolution criterion we're holding to is specific, and we should defend why we've set it here rather than somewhere else. We require an Am Law 50 firm to explicitly state in a client communication, earnings call, or press release that AI has replaced a headcount-bearing contract attorney role, with specific numbers cited. We've set this bar because the genuine signal in this forecast is institutional acknowledgment — not quiet adoption. The article notes firms are already 'restructuring hiring toward experienced laterals over new graduates.' That restructuring is almost certainly real. But it's not the same as attribution. A hiring shift is a budget decision. Public attribution is a liability and PR decision. These have different drivers, and the second one is what we're forecasting.

Now for the counterargument we can't dismiss: ABA Formal Opinion 512 (2024) on generative AI in legal practice requires attorneys to maintain competence, supervision, and ultimately accountability over AI-assisted work. State bar guidance across California, Florida, and New York has reinforced this. This isn't a future risk to our forecast — it's an existing structural constraint on the specific behavior we're predicting. A firm that publicly claims AI replaced its contract attorneys without attorney supervision is simultaneously confessing an ethics violation. This may be the single biggest reason our resolution criterion goes unmet even as adoption accelerates. We weight it heavily. It's the part of this thesis that keeps us up at night, and it's why 58% is not 70%.

So why 58% and not lower? Because client cost pressure is not abstract. The firms deploying Harvey and CoCounsel are under competitive pressure to pass savings to clients or absorb margin improvements. That pressure creates incentive for at least one firm to define the narrative on their own terms before a regulatory body or journalist does it for them. The Colorado AI Act taking effect June 30 and California's executive order on AI-driven workforce displacement are creating a disclosure environment where silence is increasingly a strategic choice rather than a default. We moved from a prior of roughly 52% — based on technical capability alone — to 58% when it became clear that multiple state-level frameworks were beginning to make attribution visible whether firms wanted it to be or not. What would move us above 70%: a bar association that affirmatively clarifies supervised AI document review satisfies ethical obligations, removing the confession-of-ethics-violation problem from public attribution. What drops us below 45%: a high-profile malpractice case linked to AI-assisted document review that triggers industry-wide withdrawal from public claims.