Picking up the firm from the previous post. The audit practice has hit its 30% engagement startup reduction. The advisory practice is two quarters into its own AI-supported workflow project, focused on client deliverable drafting and research synthesis. The tax practice is still cautious, watching the others, citing regulatory considerations whenever someone asks when tax will move. The Managing Partner has started talking about the firm's AI program in client meetings with a confidence that's mostly warranted, which is a new experience for him.
The firm has settled into a comfortable operational connection. Two practices are running real AI-supported workflows. The numbers are good. The vendor relationships are stable. Marcus has earned credibility he didn't previously have. By the metrics most firms track, the firm is in great shape on AI.
What disrupts the comfort is a question Rita asks in a partner meeting in Q3 of the second year of the audit project.
Rita's question is uncomfortably specific. The audit practice has reduced engagement startup costs by 30%. That reduction is real and measured. But she has been looking at where the saved hours are actually going, and the picture she's seeing isn't what the original project pitch suggested. The senior associate hours that were freed up by the AI tool are not being absorbed into more billable analytical work. They're being absorbed by more engagements per senior associate. The practice has been quietly raising the engagement load on staff to capture the productivity gain as revenue. The staff have noticed. Turnover in the senior associate ranks is trending up. Two of her stronger senior associates have left in the last quarter, both citing burnout in their exit conversations. Neither said anything about AI specifically. Both described the workload as having become structurally heavier in ways that don't show up in standard metrics.
Rita lays this out to the partners. The room is quiet for a moment. The Managing Partner asks what she's proposing. Rita says she isn't proposing anything yet. She's surfacing a question. The question is whether the firm has actually been doing AI transformation, or whether it's been using AI to extract more billable hours from the same staff while telling itself a different story.
Linda, who has been at the partner meeting for the workforce update portion of the agenda, takes the opening. She has, by this point, been quietly tracking workforce signals across both AI-affected practices for eighteen months. She brings out a deck she has been refining for most of that time. The deck doesn't argue for or against AI. It argues that the firm has made workflow-level changes without making role-level changes, and that the gap between the two is producing predictable consequences: differentiation among staff, attrition among the people who weren't given a chance to adapt, and an emerging skills mismatch between the work the firm is doing and the way the firm is recruiting and developing people.
The partners listen. Dana asks several pointed questions about the financial implications of what Linda is describing. The Managing Partner asks whether Linda is recommending that the firm slow down on AI adoption. Linda says she is recommending the opposite. The firm should accelerate, but it cannot continue to accelerate at the workflow level without doing the corresponding work at the role level. The firm has been doing half of the transformation. The other half is overdue.
This is the conversation that moves the firm into stage four.
The work that follows is substantially more uncomfortable than the work that produced stage three.
Stage three asked one question: what specific workflow could we redesign with AI assistance. The answer was tractable, the implementation was bounded, the gains were measurable. Stage four asks a different question: what do we want the firm to look like in three years, assuming AI capability is foundational rather than additive. That question is not bounded. It implicates partner economics, career paths, hiring strategy, training programs, compensation structures, and the partnership track itself. Every one of those conversations is harder than the one that came before it.
The firm forms a working group. Rita and Linda co-chair. Marcus is a member. The advisory practice leader joins. The tax practice leader is invited and declines politely, citing his practice's regulatory considerations. The working group's charter, drafted by Linda and refined by Rita, is to produce a recommendation within ninety days about what the firm's roles, career paths, and operational model should look like if AI capability continues to deepen at the current pace.
The working group's first month is spent mapping the current state in more honest detail than the firm has previously attempted. What does a senior associate actually do today, hour by hour, in each practice. What portions of that work are being absorbed by AI tools versus done by the human. What skills are senior associates developing in the current model, and which of those skills are still valuable in three years. The mapping work is tedious and politically sensitive. It surfaces things the firm has been avoiding looking at, including the fact that several role descriptions on the firm's website describe work that is no longer being done in the way the descriptions imply.
The second month produces the working group's central proposal. The senior associate role, as currently structured, is going to be obsolete within three years. That doesn't mean fewer senior associates. It means the work currently labeled "senior associate" will split into two distinct roles. One role focuses on AI-augmented technical review, staff who work alongside AI tools on the analytical and judgment-intensive portions of client work, supervising the AI's output and exercising professional judgment on what the AI gets right and what it gets wrong. The other role focuses on client engagement and relationship management, staff who spend less time on technical execution and more time interfacing with clients, structuring engagements, and translating between the firm's operational delivery and the client's needs. Both roles are senior-associate-equivalent in compensation and progression, but they require different skills, different development paths, and different recruiting pipelines.
The working group also recommends that the firm sunset its current model for entry-level recruiting. The traditional pipeline, which brings in undifferentiated associates and develops them into senior associates over five to seven years, was built around an apprenticeship model where junior staff learned by doing the kind of work AI is now doing. That apprenticeship model is breaking. The working group recommends that the firm begin recruiting differentially into the two emerging tracks, accepting that some current senior associates will fit cleanly into one track and others will need to be developed deliberately into one or the other.
The recommendations are not received as enthusiastically by all of the partners.
Several of the tenured partners push back. The pushback is partly substantive (the firm has been recruiting and developing people the current way for decades and has built a strong reputation doing so) and partly personal (most of the partners came up through the current model and have a hard time seeing how an alternative would produce people they would want to make partners later). The pushback is articulated diplomatically. Underneath the diplomatic surface, the resistance is real.
The Managing Partner takes a position that is more cautious than the working group recommendation but more progressive than the resistant partners would have preferred. He endorses the role split for the audit practice, where the AI work is most mature, and asks the working group to develop a phased implementation plan over eighteen months rather than the original twelve. He defers the advisory practice changes to a follow-on review. He explicitly does not require the tax practice to engage with the model yet.
Rita is not satisfied with the Managing Partner's position. She would have preferred a firm-wide commitment. She accepts the compromise because she understands the politics of partner consensus and because the audit practice moving alone gives her enough proof-of-concept latitude to demonstrate the model. Linda is more visibly disappointed. The workforce conversation she has been trying to have for nearly three years has finally happened, and the firm's response has been partial.
The audit practice implements the role split over the following year. The implementation is uneven. Some senior associates self-identify into one of the two new tracks naturally. Others resist the categorization. Several leave the firm during the transition, including one senior associate who Rita had been mentoring for partner consideration. The leavers cite a mix of reasons, mostly amounting to a sense that the firm they joined isn't the firm they're now working at.
Linda tracks the attrition carefully. The turnover during the implementation is higher than the baseline, which is expected, but it is also concentrated in a specific demographic profile: senior associates in their late twenties who came up through the traditional apprenticeship model and who feel that the AI-augmented track devalues skills they spent years developing, while the client-relationship track requires skills they weren't recruited for. The firm is, in effect, losing people who would have made strong traditional senior associates and would have made acceptable partners under the old model, in exchange for a future workforce that will be better matched to the new model. The trade is defensible. The trade is also costly in ways that the working group's original proposal didn't quite capture.
Other vantages on the firm at this stage:
The AI-augmented review specialist. A new role, two months old in the audit practice. The person filling it is a former senior associate who was on the partner track in the old model and who found that the AI-augmented work was a better fit than she had expected. Her days are spent supervising AI-generated workpaper output, identifying where the AI's pattern matching is producing false confidence, and exercising judgment on edge cases that the AI flags for human review. The work is intellectually engaging in a way her previous workflow was not. She is also aware, in a way that doesn't yet have a clear answer, that the partnership track for her new role doesn't yet exist at the firm. She is doing meaningful work without a defined career path.
A new hire. Recruited under the new model, into the client-relationship track in audit. He has a different background than the firm's traditional senior associate pool, including time in client-facing consulting before joining the firm. His ramp-up is faster than a traditional senior associate's would have been on the relationship-management portions of the work and slower on the technical portions. The firm is still calibrating how much technical depth the relationship-track role requires. The calibration will take several years to settle.
The tax practice leader's continued caution. He has watched the audit implementation with interest and reservation. The audit practice's restructuring has surfaced workforce frictions he would prefer not to take on in his own practice. The regulatory considerations he cites are real but are also, he privately admits to himself, a reason to defer a harder conversation. He is unlikely to bring the tax practice into the new model until the firm's senior leadership requires it, which may take another year.
The Managing Partner's growing understanding. He has, over the course of the role-redesign work, become substantively more sophisticated about what AI is actually doing inside the firm. The vague confidence of the previous year has been replaced by a more specific kind of understanding. He can now describe, in detail, what the firm has changed and what it still needs to change. He has also started to see the limits of the working group's recommendations. The role split solves the senior associate problem. It does not yet solve the partner problem. The partnership track was designed for people who came up doing work that AI now does. What the partnership track should look like for people who came up in either of the new tracks is a question no one has answered, and it's a question that will determine the firm's leadership pipeline for the next twenty years.
The firm has now done the work of stage four for one of its three practices. The audit practice's roles, career paths, and operational model have been substantively redesigned around AI capability rather than augmented with AI tools. The advisory practice is following on a delayed timeline. The tax practice has not yet engaged with the model.
What's been gained is real. The audit practice has both higher operational efficiency and a more defensible model for what the work actually looks like in 2026 and beyond. The roles are honest descriptions of what people are doing. The career paths reflect where the firm needs the work to go, not where the work used to be. The new hires the firm is bringing in are being recruited for the firm the firm is becoming, not the firm it used to be.
What's been lost is also real. Several capable senior associates have left during the transition. The traditional apprenticeship model, which produced many of the firm's current partners, no longer works in the audit practice. The partnership track is increasingly uncertain for people coming up through the new tracks. The trust between leadership and staff has taken a meaningful hit in audit, and rebuilding it is going to take years rather than quarters.
The firm is doing genuine systemic redesign. The redesign is also genuinely disruptive. Both things are true at once, and any honest portrait of stage four needs to show both.
The next stage of the case study picks up the firm several years later, in a state that approaches what the diagnostic post called transformational. By then, the audit practice has stabilized its new model, the advisory practice has gone through its own redesign, the tax practice has finally engaged, and the firm is doing work it could not have done before AI was foundational to its operations. The transformation is real. So is the cost of having gotten there.
— Chris