Design and marketing have gone through plenty of frameworks over the past forty years. Brand platforms, digital-first, growth, service design, performance. Most of those changes happened inside roughly the same production model. A company commissioned a position, an identity, a website, or a campaign; a specialist team spent weeks or months producing it, handed over the finished work, and the company used it for some time. Good form was expensive, and the expense itself signaled serious intent. Channels moved slowly enough to understand. Major decisions could be made at the beginning of a project and applied consistently for years.
By 2026, none of those assumptions can be taken for granted.
It was not a neural network inside a design tool that broke the model. The “AI will change design” essays arrived in 2024; the tools have already faded into the background. What matters is that two larger shifts happened at once: businesses need to adapt more often, while the cost of producing an attempt has collapsed.
The first shift shows up in money. In 2025, AI companies captured 61% of global venture funding, $258.7 billion out of $427.1 billion (OECD). That figure proves concentration of investor attention and nothing more; the interpretation is mine, built from where I sit. Outside a handful of high-growth categories, decisions take more justification, familiar budgets are more contested, and plans built on last year’s assumptions stop adding up more often. When it becomes harder to know in advance which market, product, or message will work, a single large bet grows less attractive than several cheap, testable ones. None of this logic is new. Rita McGrath and Ian MacMillan wrote it down in 1995 as discovery-driven planning, a method for launching ventures whose assumptions had not yet been proven. What changed is not the concept. What changed is the economics that make the concept binding for an ordinary company.
The second shift I can measure in our own work at Embacy. In some projects, layout that once took two to four weeks is now completed by one designer in a day, and five hours of research covers the volume of material that used to take a week. It is worth being precise about what accelerated. AI increased the throughput of preparation: in the same time we consider more sources, develop more alternatives, and arrive at a decision better informed. The decision itself did not become faster. Good analysis still takes the time it takes. Framing the problem, choosing what matters, determining what the words should actually say are no easier than they were two years ago. Production sped up by an order of magnitude. Decisions did not.
Put these together and the change becomes larger than any tool. Polished visuals wrapped around embellished copy, assembled slowly and launched for the long term, are giving way to a different kind of work: functional, clear, memorable design built around a concrete product; demonstrations instead of broad promises; positioning that places the product inside the audience’s existing picture of the world. It is built through a sequence of hypotheses, not as one grand project every three years.
By now, part of this is consensus. Every industry report of 2026 says roughly the same thing: production is cheap, judgment is the new scarcity, taste is what remains human (4As Look Ahead 2026, AI in Design Report 2026 by Designer Fund and Foundation Capital). I mostly agree, and I think the consensus stops one step short. Companies meet uncertainty in three ways: they freeze, they do the same thing faster, or they find a new way to do it. Most of what travels under “production to judgment” today is the second response, automation of an existing process that nobody redesigned. Judgment that lives only in one person’s head does not scale, and taste that cannot leave its owner dies with every handover. The interesting question of the next few years is not who has judgment. It is how judgment becomes executable, so that work done without you continues your logic instead of starting a random new one. That is the third response, and it changes the process itself.
That question is what the rest of this text is about. First how I read the landscape, then how we are answering inside our own practice. We are inside the same uncertainty as everyone else, and we are building our response through hypotheses too. Nobody knows the full answer in advance. That may be the defining feature of the period.
Who needs this, and who does not#
One client, Rafid from Yuma AI, moved the site we had built for him in Webflow into code (you can read more about this in his article on Substack). He now ships pages himself: describe the task, receive the page with copy and layout already in place. A landing page for a particular audience or a structural change after a customer conversation can be launched the same day.
Another client, a financial company with years of history, declined an AI-ready design system. They need new design materials perhaps once a year. The final deliverables from the project will serve them for a long time.
Both decisions are rational, and the second one deserves to be said plainly, because almost nobody in this discourse says it: some companies are right to opt out. The dividing line is not between progressive companies and companies that failed to understand the technology. It runs through the rhythm of the business. For some, the website is a moving surface of the product; the offer changes, segments emerge, hypotheses queue up, and production delay becomes a direct constraint on revenue. Others depend on deep human expertise, long cycles of trust, and a relatively stable proposition. Dragging them into weekly iteration simply because the tools now allow it would solve nothing.
So the useful question contains no AI at all: what limits the speed at which the company turns a market change into an action? When the answer is production, the new tools create an immediate advantage. When the answer is decision rights, alignment, and approvals, the process has to change first. No page generator can route around five people who cannot agree on what the first screen should say. Deloitte found that 48% of organizations introduced AI without redesigning the workflows around it; only 12% reported redesign at scale with a new operating model. It is easy to read that gap as laziness. I read it differently: most teams simply do not know what redesigning the process means in practice, beyond telling everyone to chat with a bot. Redesign needs a methodology. People have to be taught to work in a new way, and experiments have to live inside the workflow itself, because nobody knows in advance what the new way looks like for their team. A tool without that loop simply produces noise faster.
If what a business needs is not more materials but a shorter path from uncertainty to meaningful action, then what it buys from designers and agencies changes too. That is the second half.
What an agency sells now#
A recent project with red light therapy equipment company Luminous Labs represents a product category that is growing alongside the overall self-care trend. The client came to us for a website and a design system for a new range of full-body therapy beds. Once we unpacked the problem, it became clear the two deliverables could not work independently. A design system needs a reference point, a concrete artifact that shows how the principles work together and what level of quality counts as correct. The website could not merely display specifications; it had to express the product’s value and let people picture the real experience of using it.
The hardest call in the project was not visual. It was deciding how far the new range could drift from the master brand. Too close, and a premium product launch reads as another SKU; too far, and the company fragments into unrelated identities. We resolved it by drawing the boundary explicitly: what must remain stable across everything the company makes, and what the new line is allowed to change. The interface built into the beds became another expression of the same boundary, in a context where new features could easily make the experience overloaded and internally inconsistent. The outcome was not simply a website, an interface, and a component library. It was the logic those parts continue, written down and agreed on. Read more about the full process and what we did for Luminous Labs in Embacy’s article on Substack.
After the essay came out, Barbara Sekulovska, founder and CEO of Luminous Labs, added a perspective I had not covered, and I want to quote it as she wrote it:
“AI doesn’t just make production cheaper, it changes what the product itself can become. We’ve increasingly started treating our hardware as the entry point into a software- and protocol-driven ecosystem. The long-term differentiation isn’t the physical product anymore, but the logic, personalization and customer journey built around it. That’s why your point about making judgment executable resonated so much. I think that’s where the real competitive advantage will be. Not using AI, but encoding expertise so it can scale consistently.”
The part I keep returning to happened after handover. The client’s team moved into active product iteration and now updates the interface without us, leaning on the principles and elements of the system. This has quietly become my measure of the work: not what launched at the final presentation, but what the client can do six months after we are gone.
Rules finally have an executor#
This kind of delivery became possible for a specific reason, and it is the mechanism I would point to if I had to pick one thing that actually changed in the craft.
A PDF brand book was usually a dead archive. Applying it required someone to find the document, read it, understand it, remember the rules, and translate them correctly into a new situation. In practice, the brand’s real behavior drifted away from its written description within months. The rules existed; they just had no executor other than a busy human.
Now they can have one. When the position, message hierarchy, visual grammar, tone of voice, and boundaries of acceptable variation are stated clearly enough, embedded in components, supplied with context, and connected to a review process, a machine can apply them at the moment of production. Not without supervision, not without errors, and not without a person judging the result and updating the system. But for the first time, rules can be an executable part of the workflow rather than a reference document sitting next to it.
This is where the taste conversation usually ends and where I think the actual work begins. Writing rules was always the easy half of the job. Making rules executable, so that a page assembled without you continues your language rather than inventing a plausible new one, is considerably harder, and it is a different skill from either designing artifacts or having good taste. Manual craft does not disappear in this picture. The language is still established through canonical artifacts: the core page, the central product story, the defining visual composition, made by people at full attention. Their role changes. They stop being final states and become the beginning of a living system. A brand stops being a fixed collection of approved outputs and becomes a logic that can change with the company without losing itself.
What convinces a buyer now#
Cheap production devalued more than layout. It devalued promises. A clean website with abstract graphics and copy about “reimagining the industry” can be assembled in an evening. A buyer may not know which tool produced it, but they feel the inflation of form: too many products look equally confident for confidence itself to remain evidence.
Abstract language creates a second problem, and this one operates inside the team. When everyone discusses a “seamless experience,” everyone imagines a different product. Everyone nods. The disagreement appears only when someone finally puts a concrete screen or a working demo on the table. We keep a simple rule inside the agency for this: whenever you discuss an abstraction, show the concrete thing.
In our conversations with large B2B companies, the brief increasingly begins somewhere other than “we need a redesign.” The question is what will convince their audience. For a C-level audience my answer is usually compact: a clear product message on the first screen, specific use cases, and an immediate way to see the product working, so that nobody has to work through the entire website before discovering whether the product has anything to do with their problem. The demo does not have to be an interactive module. A screencast that literally follows an important user journey, a sequence of real screens, a static walkthrough. Here is the problem, here is the person, here is what they do in the product, here is what changes as a result.
Turning up the volume is easy. Urgency, emotion, and grandeur write themselves; AI is particularly good at producing this register when the prompt asks it to “make it powerful.” Showing the product precisely is harder. The team has to understand the audience, select the right journey, expose real details, and resist hiding limitations behind atmosphere. Calm specificity therefore becomes a signal of quality in itself. It suggests care not only in the communication but in the relationship between the product and its actual use.
None of this makes brand irrelevant. Ten competitors may demonstrate a similar user journey; one will feel like reliable infrastructure, another like a radical new operating model, a third like a lightweight tool for a small team. That difference is not created by one layer alone. Brand creates recognition, taste, and context. Positioning explains who the product is for, when it matters, and what it should be compared with. The demo shows that the promised value exists. Brand helps people remember, positioning helps them understand, the demo helps them believe; differentiation is what the three produce together. Problems begin when one layer attempts the work of the others. Branding has spent years covering the absence of product specificity with metaphor and atmosphere, and a collection of literal product walkthroughs cannot create a position or a memory on its own either.
Three tests#
A practical way to locate your company on this map.
The competitor test#
Remove the company name from the homepage. Could a direct competitor use the page unchanged, replacing only the logo? If so, everything distinctive about the product lives somewhere outside the design and the copy, and the page reveals little of it; its form can now be reproduced in an evening. If you fail this test, the next brief is positioning, and a restyle will fix nothing. We apply this test to our own work, sometimes painfully.
The untested hypotheses test#
Do not count the materials you shipped this quarter. Count the hypotheses you did not test because testing cost too much: the segment page that was postponed, the message that was never tried, the scenario that never reached the market. This is the hidden cost of the old production model; it never appears in a budget because it looks like a decision not to spend. If you count more than a few, build the testing loop before you build another beautiful site.
The subtraction test#
Take the plan for your next launch and mentally remove all production time. How much shorter does the launch become? If the answer is “hardly at all,” production is not your constraint. It sits in decisions, alignment, or the absence of a clear direction, and tools are not the place to begin. Start with who gets to decide.
What comes next#
It is too early to write the new playbook, and the genre may not return in its old form; the environment changes faster than universal recipes can stabilize. The working substitute is more modest. A position the company is genuinely prepared to hold. A set of hypotheses. Tests that distinguish a viable direction from an attractive story. A rhythm appropriate to the particular business rather than to a general theory of how every company should operate.
For us as an agency, the conclusion has gradually become clear. The model in which a client depends on the volume of our production capacity is ending, and it is not worth defending. The work increasingly has to be measured not only by what launches at the end of the project but by what the client can do after we leave: ship new materials, test hypotheses, evolve an interface, and remain inside one coherent logic while doing so.
There is one question I have not answered, and I want to end on it rather than on a conclusion. Some part of judgment survives being written down; it becomes principles, components, boundaries, a system another person or a machine can continue. Some part dies the moment you write it down, and forcing it into rules produces exactly the generic output everyone is tired of. Where that line runs seems to differ brand by brand, and I do not have a formula for finding it. Locating it, project by project, is the actual work now. I can live with not knowing yet. The profession has not had a more interesting problem in a long time.