The Brief Always Arrives Wrong
There is a demonstration that circulates in studios and lecture theatres, and it tends to land with the force of an argument even though it is only a performance. Someone opens a tool, types a sentence, and within seconds produces logo concepts, UI screens, a brand colour system with rationale attached. The implication settles without being stated: look what the machine can do; look what you are no longer needed to do. The conclusion; that designers face displacement; is treated as too obvious to require defending.
The demonstration mistakes fluency for understanding, and output for work. It shows that a system trained on the residue of past design decisions can reproduce their surface appearance quickly and cheaply. It does not show that design consists principally of that kind of production, or that the situations which call for design will become less frequent because production has become frictionless. Behind this mistake is a picture of design as a fixed inventory of things to be made, waiting in a queue for a practitioner to clear it. Automate the making, and the queue empties. What remains is a shrinking perimeter, retreating with each model release.
This picture is wrong. Working out why it fails is not a matter of professional self-reassurance; it matters because those who hold it are making decisions about hiring, education, and practice on a false account of what design actually does.
The Brief Is Where the Work Begins; and It Is Always Wrong
A striking proportion of design effort is expended not on producing artefacts but on establishing what they should be. The brief that arrives on a designer’s desk is typically a compressed and partially mistaken account of a situation the commissioning organisation does not fully understand. Designers do not receive problems. They help constitute them.
A hospital once commissioned a wayfinding system for an outpatient building where patients were routinely getting lost. The brief asked for signage. A designer who investigated; walking the building with first-time visitors, watching where they hesitated, asking staff which questions patients asked most; found that the decisive problem was not the absence of signs but the position of the reception desk, hidden behind a structural column and invisible from the entrance. The most consequential design work on that project produced no designed artefact. It produced a conversation with facilities management and a modest structural change.
This upstream work has no stable form. It does not produce something that can be evaluated by looking at it. Generative systems have no access to it because it was never documented; it lives in site visits, in conversations, in the moment a researcher watches a user hesitate for three seconds at a choice they should not have to make. That knowledge must be gathered freshly in each situation, because each situation is particular in the ways that matter.
Tools Have Always Raised the Stakes
The assumption that better tools mean less work runs against everything the history of the discipline shows. When digital pre-press arrived in the early 1990s, it made multi-page layouts producible in hours rather than weeks. The expected consequence was smaller workloads. What actually happened was that clients, now aware changes were cheap, began requesting more of them. Revision cycles multiplied. Non-designers became participants in iteration because the cost of their participation had fallen. The total labour per project did not shrink; the nature of the process changed, and the amount of work in it held or grew.
Interface design followed the same logic. When prototyping tools made clickable mockups producible without code, clients who had accepted static screens began expecting flows. Flows surfaced edge cases. Edge cases required decisions that had not been made. Unmade decisions required research that had not been done. Each increase in what could be easily produced raised the expectation of what should be demonstrated, and raised it faster than the tools advanced. The prototype that was supposed to close a conversation reliably reopened it.
Generative AI is already exhibiting this at the concept stage. The capacity to produce twenty visual directions in an afternoon does not compress the concept phase; it expands the option space to a point where clients want to explore more of it, where disagreements a narrower set would have suppressed, now become visible and require resolution. The easier the making, the harder the deciding. Deciding is what designers are for. [1]
The most consequential design decisions often produce no visible artefact. They produce clarity about what not to make, and why the stated problem is not the real one.
A Position Requires Someone to Hold It
Much of what design produces is not objects but positions; stances taken on contested questions about what a product should do, who it is for, and what it should cost in human terms. These positions are arrived at through a process that is irreducibly social: presentations, critiques, negotiations, and the particular kind of argument [2] that happens when someone with expertise in how people experience things sits across a table from someone with authority over what gets built.
A generative system can produce artefacts that embody a position without being able to hold that position under pressure. It cannot appear in a steering group and explain, when challenged, why a navigation structure was organised as it was, what alternatives were rejected, and what evidence supports the choice. It cannot push back when a marketing director insists on visual decisions that undermine usability, in a way that carries the weight of professional knowledge and personal accountability. It cannot be wrong in public; wrong with consequences, wrong in a way that requires understanding why.
When a medical device causes harm because its interface was poorly designed, the accountability structures that follow name practitioners and organisations, not tools. That assumption does not dissolve when a generative system is used in the making; it extends to cover the use of the system itself. The expansion of generative tools into design practice creates new questions of professional judgment; when to accept generated outputs, when to question them, how to identify the subtle failures of contextual fit that a system optimised for statistical plausibility reliably produces; that demand more expertise, not less.
Most of the World Has Not Yet Been Designed
The displacement argument assumes that design demand is currently being met; that the world has roughly the right amount of design, and that AI threatens the existing share of it. This premise is wrong by a wide margin.
Public services in most countries operate on interfaces and communication materials that have not been reconsidered in decades. Medical devices for clinical settings are frequently produced by engineers with no usability training. Assistive devices outside the premium segment, packaging for products used by people with low literacy, agricultural equipment for smallholder contexts, emergency communication in regional languages; in each of these domains the absence of design is not neutral. It is a failure with measurable consequences for the people who live inside it.
Design has been absent not because there is nothing to do but because the cost of doing it made it economically inaccessible relative to the budgets involved. If generative tools reduce production costs significantly, they do not reduce the need for design skill in these contexts; they potentially remove the barrier that has kept it out. Whether that opening is taken up is a question about the profession’s choices and its willingness to orient towards difficult, unglamorous, high-stakes problems; not a question about whether sufficient work exists.
Seeing Is an Argument
The version of displacement focused on visual production; AI can generate images, therefore image-makers are redundant; rests on a category error. It treats visual output as decoration applied to content that exists independently of it. Visual decisions in design are not decorative. They are arguments made in the register of perception.
The choice of a typeface for a public health leaflet is a claim about the intended reader and how they should be addressed. Using photography rather than illustration in a financial context is a claim about the relationship between institution and recipient; warmth against authority, proximity against distance. The design of a data visualization is an argument about which decisions are primary, what a user under time pressure should notice first, and what can safely recede.
Generative systems can reproduce the visual conventions associated with these decisions without modelling the argument those decisions make. An image generated for a maternal health application looks like such applications look; optimised for recognisability against a corpus, not for the specific communicative intent of this application with this population. It has no position on whether clinical authority or human warmth should lead, because it has not been told what the application needs to do to the person who opens it, and would not know how to make that intention visible in form even if it had. The judgment that translates intent into visual argument is not an addition to production. It is what makes the difference between a designed artefact and a plausible-looking one; a difference invisible in a demonstration and consequential in use.
None of this forecloses the real difficulties ahead. Disruption to specific tasks is already happening, and it falls most heavily on those still learning; people who have not yet accumulated the contextual judgment and professional relationships that constitute the less automatable parts of the work. That is a genuine problem, and pretending otherwise helps no one.
But task disruption is not professional elimination, and the confusion between the two is where the real damage will be done; not by the tools themselves, but by a misreading of what they threaten. A profession that responds by competing on production speed will lose, because it is the wrong competition. One that understands the disruption correctly; as a redistribution of value within practice, not a deletion of it; is in a position to respond with some intelligence: investing in the capacities that have always been hardest to develop and easiest to undervalue, taking seriously the vast territories of unmet need, and insisting on the difference between a machine that makes things and a practitioner who understands what should be made, for whom, and why.
The brief always arrives wrong. That is not a problem that needs solving. It is a description of where the work begins.
Notes
[1] The pattern: digitisation of production expanding revision cycles and client involvement rather than reducing overall labour – has recurred with each subsequent wave of design tooling, from desktop publishing through to parametric CAD and component-based interface design. It has not been the subject of rigorous longitudinal study across the profession, which is itself a symptom of how poorly the economics and sociology of design practice are documented compared to analogous knowledge professions.
[2] Design rationale is a synonym for argumentation. Rittel (1972) was the first to advocate systematic documentation of design rationale as part of design. He sees design problems as fundamentally open ended and controversial in the sense that there are no objective criteria for closing problem definitions and settling disagreements. Such closing and settling are necessary for design, but, for the designer, the decisions on closing and settling are judgmental and political in nature. The design rationale takes the form of a network of issues (design questions), selected and rejected answers, and arguments for and against these answers. Rittel’s framework – issues, positions, arguments – needs to be written as an essay on its own terms rather than as a reference. My next essay The Web of Issues is precisely that.