When a product manager can stand up a working prototype in Lovable over lunch and an engineer can ship a passable feature from a v0 generation before the design review even happens, the question your CEO asks in the Q3 offsite stops being rhetorical. "Why do we need a design team this size if everyone can design now?" is a fair question, and the honest answer is not "because we have taste" - that answer has been losing this argument for two years. This lesson teaches you to write the one document that wins it: a defensible, executive-ready definition of design's mandate in an AI-native product org. Not a manifesto about craft. A function-definition memo that names what shifts, what stays, and what gets harder, in language a CFO will fund and a PM will respect. By the end you will have the structure, the claims, and the worked example to publish your own.

The Question Behind the Question

Start by being honest about what is actually being asked. When a CPO says "PMs can prototype in Lovable now," they are not making a tooling observation. They are testing a hypothesis: that design was always a production bottleneck, that AI has removed the bottleneck, and that the headcount which existed to clear it is now slack the business can reclaim. If you respond to the literal sentence ("Lovable is great for exploration but not production"), you have lost, because you answered a tooling question and they asked an org-economics question. The function-definition memo exists to answer the question that was actually asked.

The hypothesis has a kernel of truth, and pretending it does not is the fastest way to lose credibility. Producing competent screens genuinely is cheaper than it was eighteen months ago. A PM genuinely can get a clickable thing in front of a customer without you. An engineer genuinely can scaffold a feature from a generated component. If your definition of design's job was "the people who make the screens," then yes, that job is being commoditized in real time and you should be worried. The memo's first move is to concede this completely and without flinching, because the concession is what earns you the right to make the harder claim that follows.

The harder claim is this: in an org where producing interfaces is nearly free, the scarce and valuable thing is no longer production. It is the judgment about which produced interface is correct - and that judgment does not get cheaper when generation gets cheaper. It gets more valuable, because there is now vastly more generated output that needs judging, and the cost of shipping the wrong generated thing is exactly as high as it ever was. Your memo is the document that relocates design's mandate from the thing that got cheap to the thing that got scarce.

What a Function-Definition Memo Is, and Is Not

A function-definition memo is a one-to-two-page document that states, in the present tense and in the language of the business, what the design function is for in this specific org at this specific moment. It is not a team charter (which is internal and aspirational), not a roadmap (which is a list of activities), and not a values statement (which a CFO ignores). It is closer to a job description for an entire function: here is what we are accountable for, here is what we are not, here is how you will know we are doing it well.

The distinction that matters most is between a memo that describes activities and one that describes accountabilities. "Design runs research, makes mocks, builds the system, and audits accessibility" is an activity list, and activity lists are exactly what get cut when activities can be automated. "Design is accountable for the correctness of every user-facing decision the product ships" is an accountability, and accountabilities survive automation because someone has to own the outcome even when a machine produced the artifact. The memo you write is built on accountabilities, and it names the activities only as evidence of how the accountability is discharged.

Why You Publish It Rather Than File It

The word "published" in the artifact is load-bearing. A memo in your drafts protects nobody. A memo you have circulated to your CPO, your peer PM lead, your engineering counterpart, and your own team - and that they have read and reacted to - becomes the shared definition the org operates from. When the offsite question comes, you do not improvise an answer; you point to a document the room has already seen. Publishing converts your private view into organizational furniture, and organizational furniture is much harder to cut than a person's opinion. The act of getting it read is most of the work, because a definition nobody agreed to is just your wish.

What Shifts

The memo's spine is three honest lists: what shifts, what stays, what gets harder. Start with what shifts, because conceding the shift first is what makes the rest believable. Be specific and be unsentimental.

Production shifts from a design activity to a shared capability. The making of a first-pass screen, a clickable prototype, a set of copy variants, a starter component - these are no longer things only designers do, and pretending otherwise insults everyone's intelligence. A PM in Figma Make, an engineer in v0, a marketer in Adobe Express: they all now produce design artifacts. Your memo names this directly. Design does not own production anymore; design owns whether the production is correct.

The designer's time reallocates from generation to judgment. If a model produces the first eighty percent of a mock in nine seconds, the designer's hours move to the last twenty - the hierarchy under unusual density, the destructive-action placement, the edge-case states, the brand distinctiveness - and to the harder meta-work of deciding which generated hypotheses to advance at all. The memo should state plainly that the value of a designer-hour is rising even as the number of hours spent pushing pixels falls, because the hours now buy correctness rather than production.

The unit of design work shifts from the artifact to the decision. A senior designer's deliverable used to be a Figma file. Increasingly it is a decision, documented: this generated direction over those two, for these reasons, with this evidence, and here is what we overrode and why. The artifact is cheap; the decision and its provenance are the durable product. Your memo names the decision as the unit, because that is what reframes design from a production cost into a judgment investment.

What Stays

The "stays" list is where you make the affirmative case, and the discipline is to name things that genuinely do not get automated rather than things you wish would not. Vague claims here ("taste stays") read as nostalgia. Specific claims survive.

Ownership of user intent stays. A generative model has no model of your user's intent because it has no user in its head; it has a probability distribution over its training data. Someone in the org must hold the question "what is this person actually trying to do, and does this interface serve it?" and check every generated thing against the answer. That someone is design. This is not a craft claim - it is a structural one, and structural claims survive automation in a way craft claims do not.

Accountability for correctness stays. When a generated screen ships a destructive action in the muscle-memory position of a safe one and a user loses their data, "the model produced it" is not a defensible answer to the postmortem. Someone is accountable for the correctness of what shipped to a human being. Your memo claims that accountability explicitly for design, because an accountability nobody owns is an accountability the business cannot trust, and an unowned correctness function is a liability waiting to surface.

Brand distinctiveness stays. Generative models converge - on the same dramatic lighting, the same centered composition, the same competent-but-anonymous interface patterns. An org that lets a year of generated assets accumulate without a custodian of distinctiveness wakes up looking like every competitor that used the same models. The memo names design as the function that resists convergence, that encodes what makes this product look and feel like itself rather than like the average of its category. This is a CMO-relevant claim, which means it has a budget line behind it.

Cross-surface and edge-case integrity stays. Models produce one canvas at a time and design the happy path. The coherence of a flow across surfaces, the integrity of the error and empty and loading states, the responsive behavior at the breakpoint nobody generated - these require holding the whole system in a head, and the holding is human. Your memo names integrity-across-the-system as a standing accountability.

Concede that production got cheap. Then claim the thing that got scarce: not the making of interfaces, but the judgment about which interface is correct - judgment that grows more valuable the more cheap generation there is to judge.

What Gets Harder

The "gets harder" list is the one most leaders omit, and the one that earns the most trust, because naming your own new difficulties signals that you are reasoning honestly rather than defending a fiefdom. A memo with no "harder" list reads as a sales pitch.

Verification becomes a standing tax. Every generated artifact that enters the pipeline carries a verification cost - someone has to check the destructive actions, the states, the responsive behavior, the brand fit, the accessibility. As the volume of generated output rises, the verification load rises with it, and it does not feel like progress to the people doing it. The memo names verification as real work that needs staffing and tooling, not a thing that happens for free in the margins. Leaders who pretend AI removes work, rather than relocating it to verification, will be blindsided when their team burns out checking machine output.

Maintaining a defensible boundary gets harder. When everyone can produce design artifacts, the line between "a PM explored an idea" and "a PM shipped an unreviewed interface to a customer" blurs constantly, and the blur is where products break. Holding that boundary - exploration is welcome, shipping unreviewed generated work to users is not - becomes a continuous diplomatic and political task rather than a settled fact. The memo names this and proposes the governance to hold it, because a boundary nobody defends erodes by default.

Protecting craft development gets harder. If juniors never produce the first eighty percent because the model does, where do they build the judgment to evaluate the last twenty? The apprenticeship path that produced senior judgment is exactly the path AI shortcuts. The memo names the development of judgment as an explicit, deliberate program rather than a byproduct of doing the work, because the work that built judgment is now done by a machine and the judgment will not appear on its own.

A Worked Example: Drafting the Memo for a 30-Person Org

Make this concrete. Imagine you lead design for a Series C product company - thirty designers across six product squads, a two-person design-systems team, a brand designer, and a research lead. The CPO has just watched a PM demo a Lovable prototype in a leadership meeting and has scheduled a "design org sizing" conversation. Here is how the memo takes shape, section by section, so you can see the moves rather than just the principles.

The opening concession. "Producing competent interfaces is now substantially cheaper than it was in 2024. Product managers can prototype in Figma Make and Lovable; engineers can scaffold from v0. This is real, and our function definition reflects it: design no longer owns the production of interfaces. We own whether the interfaces we ship are correct." Two sentences in, you have agreed with the CPO's premise, which means they are now reading the rest as a fellow realist rather than a defensive incumbent.

The accountability claim. "Design is accountable for the correctness of every user-facing decision this product ships: that the hierarchy serves the user's intent, that destructive actions are safe, that the experience holds across surfaces and edge cases, that the brand stays distinct, and that what we ship meets WCAG 2.2 and our legal obligations. We discharge this accountability at the volume AI now makes possible - which is higher, not lower, than before, because there is more generated output to judge." Notice the move: the volume argument turns the CPO's efficiency frame in your favor. More generation means more judging, which means the function's load went up, not down.

The explicit boundary. "We are not the people who make the screens; anyone can make a screen. We are not blocking exploration; PMs and engineers prototyping is a feature of an AI-native org, not a problem. We are the gate between 'someone generated this' and 'we shipped this to a customer,' and we are accountable for what passes that gate." This is the sentence that prevents the CPO from concluding that design and PM-prototyping are redundant. They are not redundant; they are different functions, and the memo says which is which.

The honest difficulty. "This shift raises three costs we are managing explicitly: a rising verification load as generated volume grows, a continuous need to hold the line between exploration and unreviewed shipping, and a deliberate program to develop junior judgment now that the work which used to build it is automated. We are not asking to grow headcount; we are asking to retool the headcount we have toward judgment, verification, and the design system that makes both scale." The ask at the end is precise and modest, which is what makes it fundable. You did not ask for more people. You asked to keep the people and change what they do.

That memo fits on a page and a half. It concedes everything the CPO believes, claims the thing that actually matters, draws the boundary that prevents the redundancy conclusion, and names its own difficulties so honestly that the document reads as reasoning rather than lobbying. When the offsite question comes, you do not improvise. You point.

The Three Ways the Memo Fails

Most function-definition memos fail in one of three predictable ways, and naming them is the fastest way to write a memo that does not.

The craft-nostalgia failure. The memo argues that design matters because designers have taste, sensitivity, an eye. This loses every time, because taste is unfalsifiable to a CFO and sounds like a luxury in a year of layoffs. The fix is to never make a craft claim where a structural claim is available. "Designers have taste" loses; "someone must own whether the generated interface serves the user's intent, and that ownership does not get cheaper when generation does" wins, because it is a claim about how value works rather than how designers feel.

The denial failure. The memo minimizes what AI can do ("Lovable prototypes are not real," "v0 output is not production-grade"). Even where this is partly true, it reads as denial, and an executive who suspects you are in denial about AI will reallocate around you. The fix is the opening concession: concede more than the CPO expects, then make the harder claim. Conceding is a position of strength, not weakness, because it earns you the credibility to be believed on the part that matters.

The everything failure. The memo claims design owns research and production and prototyping and engineering-handoff and brand and strategy and everything else, with no "not." This reads as empire-building, and a CFO funds focus, not empires. The fix is a strong, explicit list of what design is not doing and not owning. A function with no boundary has no strategy, and a memo with no "not" has no spine.

Putting It to Work This Quarter

Draft the memo this week, before the sizing conversation finds you. Write the three lists first - what shifts, what stays, what gets harder - in the most unsentimental language you can manage, and pressure-test every "stays" claim by asking whether it is structural or merely nostalgic; cut the nostalgic ones. Then write the opening concession so generously that a skeptic reads on, the accountability claim so specifically that it could be measured, and the boundary so explicitly that no one concludes design is redundant with PM-prototyping. End with an ask that is precise and modest, because a modest ask backed by a strong argument gets funded and a grand ask backed by a weak one does not.

Then publish it. Circulate it to your CPO, your peer leads in product and engineering, and your own team. Ask each of them to react, and incorporate the reactions, because a definition the room helped write is a definition the room will defend. The deliverable of this lesson is not a clever document; it is a shared, organizational answer to the question your CEO is going to ask, written down before the question arrives, so that design's mandate in your AI-native org is a settled fact rather than a thing you defend on your heels in a room that has already half-decided. The function gets defined by someone. The memo makes sure it is defined by you.

Key Takeaways

• When PMs prototype in Lovable and engineers ship from v0, the real question is not about tooling but about org economics: why fund a design team this size if production got cheap. The function-definition memo answers the question that was actually asked.
• Concede first, completely: producing interfaces genuinely got cheaper. The concession earns the right to make the harder claim - that judgment about which interface is correct got scarcer and more valuable, because there is more generated output to judge and the cost of shipping the wrong thing is unchanged.
• Build the memo on accountabilities, not activities. Activity lists ("design makes mocks") get cut when activities automate; accountabilities ("design owns the correctness of every user-facing decision") survive, because someone must own the outcome even when a machine made the artifact.
• State three honest lists. What shifts: production becomes a shared capability, time moves to judgment, the unit becomes the decision. What stays: ownership of user intent, accountability for correctness, brand distinctiveness, cross-surface integrity. What gets harder: verification as a standing tax, holding the exploration-versus-shipping boundary, developing junior judgment.
• Avoid the three failure modes: craft-nostalgia (replace every taste claim with a structural one), denial (concede more than expected), and the everything-claim (a memo with no explicit "not" has no spine and reads as empire-building).
• Publish it, do not file it. A circulated, reacted-to, agreed-upon definition becomes organizational furniture that is hard to cut; a memo in your drafts protects no one. Define the function before the offsite question finds you, so the answer is a document the room has already seen rather than an improvisation under pressure.