What a Director, Head, or VP of AI owns that a CTO doesn't, current Australian comp bands for each title, and why almost every one of these hires is off-market.
A Director of AI, Head of AI, or VP of AI is a distinct mandate from a CTO or VP Engineering. The seat owns the AI and machine learning roadmap, and often the pipeline that turns research into a shipped product, not general engineering delivery. This guide is for founders who already have their first ML engineer and now need someone who owns the strategy, not just the code.
Most founders reach for this hire the same way they reached for their first VP Engineering: by writing a job description that mostly restates the engineer they already have, with a bigger title attached. That doesn't work here, because the job the title is trying to describe barely existed five years ago. At Perpetuate, Stephen Tung runs these searches inside our Executive practice, and this piece covers what the role actually owns, the sector patterns we're watching right now, what it costs in Australia today, and why almost every AI leadership search like this happens off-market before an internal team ever hears the person's name.
Three or four years ago, AI was a feature. A company had a product, and somewhere inside engineering sat one or two people building a recommendation model, a fraud score, or a churn predictor that improved a metric the product already tracked. The person running that work reported into engineering, because the work was engineering: a model was one more service behind an API, evaluated the same way any other service was evaluated.
That arrangement stopped matching reality once the product itself became the AI. A growing share of companies now ship something where the AI isn't a feature bolted onto a stable product. It is the product: an agent that does the work a person used to do, a computer vision pipeline running on hardware in the field, an orchestration layer stitching together several models into one workflow a customer pays for directly. When the AI is the product, the technical risk moves with it. The hardest call a company makes is no longer an infrastructure decision an experienced VP Engineering can own competently. It is a model decision: which architecture, which foundation model to build on or around, how much to fine-tune versus prompt, what to build from scratch because nothing on the market solves the specific problem.
The Director of AI title didn't exist five years ago because the job didn't exist. Now the hardest technical bet at a growing number of startups is a model decision, not an infrastructure one.
Perpetuate Talent, on why this role exists
Most founders are hiring for this title for the first time, which is exactly when it gets mistaken for a rename of Head of Data Science, or for a CTO role with an AI-flavoured job spec bolted on. Neither guess matches what the seat actually spends its week doing.
A CTO still owns the company's overall technical bet: architecture, platform, build-versus-buy across the whole stack, and the standing to defend that bet to a board. A VP Engineering owns the delivery machine: hiring plan, sprint cadence, on-call, the org that ships against a direction someone else set. Neither role is built to own what a Director, Head, or VP of AI owns specifically:
At a company past its first ML engineer, that's the gap. The CTO is busy owning the rest of the stack. The VP Engineering, where one exists, is busy running delivery. Nobody currently on the team is spending their week deciding what the AI roadmap should be and building the team to execute it. That's the seat sitting empty right when it starts costing you.
Three patterns show up consistently across the searches landing on Stephen's desk right now, and each one is a reasonable trigger for this hire on its own.
Post-raise team-building at scale. A funding round that lets a company go from two or three people touching AI to a genuine team of ten or more, often before the product itself has fully proven out, is one of the clearest signals we see. Someone has to own that build before the headcount arrives, not after.
Physical AI and edge-AI hardware convergence. Computer vision running on hardware in the field, biosensors, edge compute on solar-powered devices: wherever AI has to survive outside a data centre, the leadership gap widens, because the skill set spans model work and hardware constraints most engineering leaders have never had to reconcile.
LLM and orchestration-heavy product surfaces. Once a meaningful share of the product is an agent, a copilot, or a chain of models doing real work rather than a single model sitting behind an API, the roadmap questions multiply faster than a CTO stretched across the whole stack can keep up with.
The specific companies vary, but four sectors keep coming back across the AI leadership mandates open right now. Names are withheld for client confidentiality; the shape of each mandate is real and current as of September 2026.
None of these are large companies by headcount. All of them are running AI leadership searches at a stage most conventional wisdom would call too early. The pattern says otherwise: the earlier the AI is the product, the earlier this hire needs to exist.
Count the people in Australia who have actually taken a model from a research idea to something running reliably in production, inside a domain that matters to your product, and the number gets small fast. Most of them are not looking. They were promoted into the role they're in within the last year, they're deep inside a research group or an applied AI team at a company that already trusts them, and their LinkedIn hasn't changed in eighteen months because there's been no reason to touch it.
A job ad reaches the people comparing options. It does not reach the person three companies would want and none of them can name, because that person isn't comparing anything. Finding them takes the same concentric-circle method behind every search we run, what we call the Talent System™: start from who's already known to be doing this work well, and work outward through who they'd vouch for by name. Our piece on the Talent System™ goes deeper into how that outreach actually runs.
As of September 2026, here's roughly what the three tiers of AI leadership cost in the Australian market, based on live mandates rather than published salary surveys. Figures include superannuation and ESOP or equity where noted; none of them include the premium some companies are already paying to close a candidate against a competing offer, which in this category is increasingly common.
Director of AI
A$220K–300K + super + ESOP
Execution-heavy. Usually reports to an existing CTO and owns delivery of an AI roadmap someone else has already set the direction for.
Head of AI
A$250K–270K + equity
Often the company's first AI leadership hire. Mid-scale mandate: sets direction and builds the team, without yet sitting at the same table as the CTO.
VP of AI
A$250K–300K + ESOP, up to A$400K
Strategic and peer to the CTO. The band widens sharply for genuine research leadership on pre-funding or post-raise foundational AI teams.
The comp band alone won't tell you which of these you're hiring for. A Director of AI hired to execute against someone else's roadmap and a VP of AI hired to set the research direction from scratch are different searches even when the comp bands overlap, and treating them as the same conversation is how founders end up interviewing the wrong shortlist for the title they wrote.
Signs You Need AI Leadership, Not Another ML Engineer
Director of AI, Head of AI, and VP of AI searches run through Stephen Tung's Executive practice, the same practice that places CTOs and VPs of Engineering. If the mandate is really product-side, an AI Automation Lead building AI into workflows the product team already owns rather than owning the model strategy itself, that's usually a search co-founder Sophia Philippou runs inside the Product practice instead. Worth checking which one you're actually writing a brief for before the search starts; our guide on choosing between a CTO and a VP Engineering covers the same mistake in a different mandate, and it's worth the five minutes either way.
A Director of AI typically executes against an AI roadmap someone else, often the CTO, has already set, and reports into that person. A VP of AI sets the roadmap itself and sits as a peer to the CTO rather than beneath them. Titles get used loosely across the market, so the real test in any brief is who owns the model-strategy decision, whatever the word on the org chart says.
Not always. A Head of AI role is usually the company's first dedicated AI leadership hire, and the mandate is as much about building a team and shipping reliably as it is about research depth. Foundational or post-transformer research leadership proper is a narrower, separate search, and it's the one where a research background becomes close to non-negotiable.
In almost every mandate we run, this hire reports to the CTO or sits alongside them, not instead of them. The CTO still owns the company's overall technical bet across the stack; the AI leadership hire owns the specific bet on models, research direction, and the AI team. Framing it as a replacement is usually a sign the brief hasn't separated the two mandates yet.
Perpetuate Talent asks for one hour of a founder's time to build the profile and the pitch. Two to three candidates arrive within five to ten days after that. That's time to a shortlist you can start interviewing, not time to a signed offer, and for a hire this scarce, the outreach itself is usually the slowest part of any process that skips it.
Almost always, at the seniority these searches run at. The people who've done that, taking AI from research to production, aren't comparing job offers, and a public listing mostly reaches people who are. A handful of Director of AI hires do come from an active search, usually candidates between roles after an acquisition or a shutdown, but treating that as the default plan is how most searches in this category stall.
If you're scoping this hire now, Stephen Tung takes it from here inside the Executive practice. Message Stephen on LinkedIn → and bring the honest version of the mandate, not the flattering one. If the roadmap question underneath it all is really "CTO or VP Engineering," not AI leadership specifically, our decision guide on that split is the better place to start.