The apprenticeship layer of the legal profession is the product

Perplexity's Computer for Counsel, launched on 24 June, is being sold as a workflow tool. It is also, by design, a compression mechanism aimed at the specific billable hours that fund a junior lawyer's training. The interesting question is not whether the tool works. It is whose hours get smaller, and what that does to a profession that has always trained its next generation on those same hours.

What the product actually targets. Computer for Counsel plugs into Microsoft Word, SharePoint, Outlook, Teams, Box, Carta, and Clio, and routes work through Midpage for case-law citations and LegalZoom for templates.¹ It picks among more than twenty frontier models for the user. The integration list is a precise map of where junior associates spend their billable time: drafting in Word, pulling files from SharePoint, checking citations, building documents from templates, triaging contracts in Box.

The Windows Forum analysis of the launch was unusually direct about this. It described the targeted work as "review-heavy junior work" and noted the product applies "pressure on tasks clients already suspected were inefficient."² That is not a marketing framing. It is a description of which billing lines are now contested.

Who actually loses. A first-year associate at a US transactional firm bills, very roughly, between 1,800 and 2,000 hours a year. A meaningful share of those hours, in the first two or three years, sits in exactly the work Computer for Counsel is built to compress: document review, first-pass drafting, citator checks, template assembly, due-diligence triage. This is not a peripheral category. It is the apprenticeship layer of the profession. It is how lawyers learn to read a contract, spot a risk, hold a deal in their head.

Gunderson Dettmer, the Silicon Valley transactional firm that has run Perplexity Enterprise firmwide since 2025, has now committed to deepen its use through Computer for Counsel.³ BigLaw firms of that kind have historically absorbed efficiency tools by raising throughput per associate rather than cutting associate counts. They might do that again. Even if they do, the texture of associate work changes: more review of AI output, less reading from scratch, fewer of the slow grinding hours in which judgement is built.

The mid-market is where the cushion runs out. The more consequential part of Perplexity's integration list is Clio and LegalZoom. Clio's customer base is documented as primarily solo practitioners and small firms. LegalZoom serves consumers and small businesses. These are not firms with a layer of associates to redeploy. They are firms where the lawyer is the work, or where one paralegal supports two partners, or where a contract review used to mean an hour of paid time and now means a Perplexity session.

When a tool reaches the Clio segment, the displacement question is not "will the firm hire one fewer associate next year." It is "will the paralegal's hours be cut," or "will the small firm down the road still be able to charge for the work that the LegalZoom-plus-Perplexity stack now does for a flat fee." The labour market in this segment has been compressing since LegalZoom started selling templates in the early 2000s. Computer for Counsel is the next turn of that screw, not the first.

Who benefits, and who pays the bill. The clients benefit. Clients have suspected for years that some review-heavy tasks were billed inefficiently, and they were often right. The AI vendors benefit, capturing a share of what used to be associate salary as software revenue. Partners and senior counsel, whose value is relational and strategic, are largely insulated. The associates and paralegals carry the compression.

This is the distributional pattern that recurs in professional services every time a layer of routine work gets automated. The efficiency gain accrues upward to clients and outward to vendors. The income loss lands on the people whose work was being compressed. The senior practitioners narrate the change as modernisation. None of this requires anyone to be acting in bad faith. It is just how the incidence works when the tool is sold to the people who buy the labour, not to the people who sell it.

The model-abstraction question nobody is asking. Perplexity's pitch includes the line that it picks the best of more than twenty models so the lawyer does not have to. Framed as user benefit, this is real. A solo practitioner is genuinely better off not having to evaluate which frontier model is best for a contract review on a Tuesday morning.

Framed as a structural change, it is something else. The lawyer carries professional responsibility for the output. Bar associations regulate the lawyer. Malpractice insurance prices the lawyer's risk. But the input layer, the choice of which model handles which task, sits with Perplexity, behind no professional transparency requirement and no oversight that maps onto the legal profession's existing accountability structures. The lawyer is liable for output produced by a stack they cannot inspect. That is a quiet but consequential rearrangement of where risk sits, and the profession has not yet decided what to do about it.

The harder counter-case. I want to take the access-to-justice argument seriously, because it is the strongest case for this kind of tool. There is a real population of people and small businesses who currently go unrepresented because the cost of a lawyer's hour is higher than the value of their dispute. If Computer for Counsel and its peers materially lower the floor on what routine legal work costs, some of that gap closes. That is a genuine human-consequences gain and it should not be waved away.

The question is whether the cost reductions reach the underserved client, or whether they are captured as margin by the platforms in the middle. The honest answer is that we do not yet know. LegalZoom's history is mixed: real consumer access, alongside steady pressure on the paralegal labour market that used to do similar work for a wage. Computer for Counsel is likely to repeat that pattern at a larger scale, with the gains and the losses landing on different people.

What I am confident about is this. The legal AI conversation in 2026 is still mostly framed as a contest between vendors: Perplexity versus Harvey versus Microsoft versus Anthropic versus OpenAI versus Palantir. That framing is useful for the vendors and unhelpful for everyone else. The story the profession should be paying attention to is not which company wins the legal vertical. It is what happens to the people whose first three years of work just became the product.

Footnotes

¶