A regional IT services firm — the kind that sells large-scale transformation programs to government and enterprise clients — is hiring a Practice Unit Leader for Data & AI. Director-level. Fifteen-plus years of experience. Six years leading a managed services business. Deep expertise in sovereign architecture, multi-cloud strategy, AI governance, and secure AI deployment. If this company builds and sells AI practices for a living, why does it need a human to lead its own?
The vacancy functions as a confession. The job description describes a person who can architect an AI practice, commercialize it, govern it, and run it at scale. That same job description outlines, with uncanny precision, the exact set of tasks an AI system can now perform for the practice itself. The firm is hiring someone to sell AI transformation while implicitly confirming that its own transformation hasn't happened yet.
So let's do them a favor and design the replacement.
What the Practice Unit Leader Actually Does
Strip away the consultant vocabulary and the role has four pillars. First, offering development — packaging reusable solution frameworks, accelerators, and reference architectures that delivery teams can sell repeatedly. Second, delivery oversight — running the managed services business: keeping utilization high, margins intact, and clients from escalating. Third, governance and compliance — ensuring that what the teams build across public and private cloud satisfies data sovereignty, cybersecurity controls, and the increasingly loud demands of AI governance frameworks. Fourth, commercial leadership — turning technical capability into revenue, whether by shaping bids, building go-to-market motions, or convincing a government CIO that this firm's data platform beats the one pitched by the competitor who showed up yesterday.
Each of those pillars is now partially or fully automatable — in practice, today, with tools that already exist.
How AI Takes Each Pillar
Offering development goes first. A Practice Unit Leader today spends a meaningful chunk of time maintaining slide decks, reference architectures, and accelerator catalogs — the reusable IP that makes a consulting firm more than a body shop. An AI system can ingest the firm's entire corpus of past proposals, delivery documentation, and architecture decisions, then generate new offering packages on demand. You want a "Sovereign AI Platform on a major public cloud for Public Sector Healthcare"? Prompt it. The system produces a thirty-page solution brief, a reference architecture diagram, a cost model, a risk register, and a list of the seven previous engagements where something similar was delivered. It draws from your institutional memory better than any individual director could, because no director remembers every project. The AI does.
The reusable frameworks become genuinely reusable when a model can reassemble them for each new client context without a human spending two days in PowerPoint.
Delivery oversight is the pillar that sounds most human — managing people, reading the room, calming angry clients. But peel it back. What does a delivery leader actually do day to day? They review status reports, flag slipping timelines, escalate risks, approve resource assignments, and sit in steering committee meetings saying things that could have been an email. An AI system connected to the firm's PSA platform, project tracking tools, and financials can monitor every active engagement continuously. It sees the utilization dip before the human does. It notices that the senior architect on the government data migration project hasn't logged billable hours in four days and sends a nudge. It drafts the steering committee deck, pre-populates the risks, and writes the carefully worded email to the client about the timeline adjustment.
The human delivery lead's remaining job — the actual conversation with the upset client — shrinks to maybe two hours a week. The rest was information processing, and information processing is not a job anymore.
Governance and compliance is the most rule-bound pillar, because this is the pillar where firms charge the most and where the work is most rule-bound. Data sovereignty frameworks, AI governance controls, cybersecurity-aligned architecture reviews — these are exercises in checking designs against large, complex, evolving rulebooks. That is exactly what language models are good at. Feed the model the relevant regulatory framework, feed it the proposed architecture, and it produces a compliance assessment in minutes. It flags the data residency issue in the cloud region selection. It catches the model-logging gap in the AI governance checklist. It generates the remediation plan. A governance team of five becomes a governance team of one human reviewer who reads the AI's output and signs off — and even that human is mostly there because someone needs a name on the document.
Commercial leadership is the last to fall, but it falls. Bid shaping, proposal writing, competitive positioning — this is now generative AI territory. The system reads the RFP, cross-references the firm's win-loss history on similar bids, drafts the technical response, prices it based on historical margins, and produces a go/no-go recommendation. The director who used to spend three weeks on a bid response now spends an afternoon reviewing and editing the AI's draft.
The Pipeline
The actual implementation is less dramatic than it sounds. You stand up an internal platform — the firm's existing cloud infrastructure, a few fine-tuned models on top of the proposal and delivery corpus, and integrations into the PSA, CRM, and project management tools. You build agents for each pillar:
| Agent | What it does |
|---|---|
| Offering Agent | Maintains and generates the IP catalog from past project data |
| Delivery Agent | Monitors engagement health, utilization, and risk flags |
| Governance Agent | Runs compliance checks against regulatory frameworks |
| Commercial Agent | Handles bid responses, pricing, and go/no-go calls |
A single dashboard surfaces what the agents flag. One human — call them a Practice Operations Manager if you want to be kind — reviews exceptions.
The fifteen-year veteran who commands a premium salary, equity, and a team of forty becomes one person reviewing a dashboard. The team of forty shrinks to maybe eight delivery consultants who do the client-facing work the AI can't do, and even they are augmented to the point where each one does what three used to.
Where It Breaks
Without this section the rest is just a sales pitch.
The thing that does not automate cleanly is trust transfer. Government clients in particular — and this firm's job description makes clear that government is a core market — buy from people they have met, eaten dinner with, and watched think on their feet in a room full of skeptical stakeholders. They do not buy AI-generated proposals from a firm that shows up with fewer senior humans. The Practice Unit Leader is, in many deals, the human the client is actually paying for — the person whose career credibility becomes the client's political cover when the transformation goes sideways. An AI system can write a better architecture document than most directors, but it cannot stand in front of a deputy minister and absorb the blame with the quiet gravitas of someone who has done this for twenty years.
There is also the sovereignty paradox. A firm selling sovereign AI architecture to government clients cannot easily run its own practice on a US-hosted foundation model without raising questions. The AI that would replace the Practice Unit Leader may itself need to be deployed in a sovereign environment to be credible to the clients the practice serves. The replacement technology has to pass the same governance bar the Practice Unit Leader enforces.
And there is the sales ritual. Consulting firms sell hours and headcount. The commercial model depends on putting senior people on engagements and billing for their time. An AI that reduces the senior headcount needed on each engagement also reduces the revenue. The Practice Unit Leader's replacement is technically feasible; commercially, it cannibalizes the business model. Firms know this, which is why they hire directors to lead AI practices rather than simply building AI practices to replace their directors.
The Verdict
The job posting asks for a fifteen-year résumé to fill a role that could be sixty percent automated by a system the hire themselves would be expected to build. The firm wants someone to lead an AI practice who will not, apparently, apply AI to the practice.
That is not a hiring decision. That is a company betting its clients are slower to notice than its own job description would suggest.
