I want to start by saying what I am and am not. I’m a working product designer at a health-tech company. I have never run a design agency. I have friends who run them, I’ve worked with a few as a client, and I read enough of Brand New and It’s Nice That to know who’s putting out interesting work. So this is an observer’s piece, not an operator’s. Take it accordingly.
If you actually run an agency and any of this reads sideways from your day, please let me know. I’d rather correct the piece than be wrong in public.
What I’m trying to make sense of is the strangest piece of design industry weather I’ve watched in eight years. Through the back half of 2025 and into this month, design agencies and studios were quietly reshaping themselves around AI without saying much publicly about it. New positioning paragraphs appeared on websites. Service pages got rewritten. Old engagement templates went on the shelf. A few well-known shops let people go without making it a story. Others doubled their rates and shrunk their team. The aggregate motion was loud even when no individual move was.
And the moves don’t all point the same direction. Some agencies bet on going more strategic. Some bet on becoming AI implementation partners. Some are betting on craft you can’t fake yet. Some don’t seem to be betting on anything, and those are the ones I’m worried about.
The economic pressure
The math, honestly, is not hard. If a competent team using Claude Code, v0, Lovable, Bolt, and a senior designer with taste can produce a credible visual identity exploration in a day, what is a $50K visual identity engagement defending?
A few things, still. The brand strategy underneath. The taste of the team picking which directions to push. The ability to navigate a difficult founder through six rounds of subjective feedback without losing the thread. The institutional patience to push past the obvious answer to the right one. The reputation that gives the work credibility when it ships.
But the production layer of that engagement, the actual hours of execution against the strategic frame, is where a lot of agency margin used to live, and that layer collapsed in 2025. Not to zero. But to maybe 30% of what it used to take, if your team is set up for it. Less if the client is willing to live with second drafts.
That collapse is the pressure agencies are responding to. Three responses are visible in the market right now.
Response one, premium strategy-only
A handful of agencies have moved up-market into pure strategic positioning. They produce less. They charge more. The deliverable is a brand strategy document, a positioning framework, an architecture of names and audiences. Visual execution is either client-side or outsourced or very narrow.
Koto is the cleanest public example I can point at. Their work has always leaned strategic, but the recent engagements I’ve seen them publish (or that founders have shared with me privately) are weighted heavily toward the strategy side, with execution that’s tight and considered but not the centerpiece. The pitch is, you can generate a logo in any color you want, but you can’t generate the right brand strategy for your particular company at this particular moment. They do that.
The work that follows in this category looks more like the kind of thing Red Antler was doing in 2018 and 2019, before they got famous for execution. Strategic identity work, very condensed delivery, very expensive engagements. The team size shrinks. The fees per project go up. The portfolio looks less full but each entry carries more weight.
This response works for agencies with established reputations. It works less well for younger shops, because the strategic positioning depends on the founder’s pre-existing credibility. You can’t market yourself into being the strategic firm. You either are or you aren’t.
Response two, AI implementation partners
A new category appeared in 2025 that didn’t exist eighteen months earlier. The AI implementation partner. These shops position themselves as the people who help product and design teams actually integrate AI tools into their workflows. They run audits. They build internal prompt libraries. They train teams on Claude and Cursor and v0. They write company-specific style guides for AI-generated content. They build internal evaluation rubrics for generated work.
Nord Projects has shaded in this direction. So have several smaller shops I’ve talked to that I won’t name because they haven’t publicly repositioned yet. The work looks like consulting, but the consulting is technical and design-flavored rather than McKinsey-flavored. The deliverable is a more capable internal team rather than an artifact.
The economics here are interesting. The engagements are usually shorter than a brand identity, but they recur. A team that pays $40K to get their AI-design workflow set up will often pay $15K a quarter after that for ongoing tuning, training, and adjustments as the tools change. Total revenue per client over two years can match or exceed a traditional identity engagement, with less concentration risk because no single project carries the firm.
The challenge with this response is that it requires real technical depth. You can’t fake your way through helping a team integrate Claude Code into their workflow if you’ve never actually shipped anything with it yourself. The agencies that are succeeding here all have at least one principal who’s a hands-on practitioner, not just a strategist.
Response three, craft AI can’t yet do
The third response is to narrow the craft position. To say, we do the kind of work the generative tools still can’t touch. Hand-drawn illustration. Custom typography. Editorial layouts that require sustained attention. Motion design with nuanced timing. Brand systems with deep historical research underneath. Packaging with physical materials.
Pentagram has always operated here, though I’d argue they’ve been quieter about it recently in a way that suggests they know the position needs to be defended actively. Manual and a number of editorial-leaning shops have leaned into the craft narrative. The pitch is, AI can produce competent visual work, but the work that wins awards, gets press, ages well, and shapes a category requires something that’s not yet on the menu of any generative tool.
This response works if the craft claim is real. The risk is that it’s getting harder to defend year over year. Generative tools improved more in 2025 than I expected, and the gap between “AI can do it” and “AI can do it well” closed in several categories craft shops were betting on. Custom typography is closer to fully generatable than it was twelve months ago. Editorial layout is mostly there for anything other than the most ambitious print work. Hand illustration still feels safe but generative image tools keep getting more controllable.
I’d genuinely bet on this category at the very high end and worry about it in the middle. If you’re in the top decile of craft on a given dimension, you have years of runway. If you’re a competent craft shop without a particular peak, the floor is rising fast.
The in-house angle
The piece a lot of agency conversations miss. In-house design teams absorbed a meaningful amount of work that used to go to agencies. Not because in-house teams got bigger. Because they got more capable at the kinds of work agencies used to be hired for.
A product company that needed a marketing site refresh in 2022 hired an agency. In 2025, the same company gave their senior product designer a week with Framer and Claude Design and got a result that was 70% as good for 5% of the cost. The remaining 30% was strategic positioning, which they handled internally because they already had the context.
That shift is hard to measure but it’s real. Several agency principals I’ve talked to mentioned, separately and unprompted, that pitches got harder in the second half of 2025 because clients increasingly have an option that didn’t exist before. Just have the team do it. The agency wins now have to clear a higher bar. “We’d never get to it internally” is no longer a sufficient reason.
The agencies that look healthiest going into 2026 are the ones that were already strategy-first. The ones in trouble are the ones whose differentiation was production speed and polish.
What I’m watching
A few threads I’m following.
How agency hiring patterns shift. The job listings I’ve watched go up over the last quarter skew toward strategy, content, and AI-fluent generalists. Production roles, the junior designers whose primary value-add was execution speed, are noticeably scarcer.
Whether new agency formats emerge that are AI-native from day one. I keep expecting to see a small studio launch with two principals, no juniors, an explicit policy of using Claude Code and v0 and Figma Make for all production, and a fee structure that reflects the AI-driven efficiency. I haven’t seen one yet that’s also good at the strategy and taste layers, but I don’t think it’s far off.
Whether the AI implementation partner category consolidates or fragments. Right now it’s a few dozen small shops doing variations on the theme. By the end of 2026, either one or two will have grown into the category leader, or it’ll have stayed cottage and gotten absorbed into the broader management consulting market.
What happens to the middle. The agencies in the strongest position are the very strategic ones at the top and the very crafted ones at the top. The agencies in the weakest position are competent generalists in the middle. That middle has been the bulk of the market for a long time. If it really hollows out, the industry shape changes in a way that affects how junior designers find their first jobs, how brands get built, and how design culture propagates. That last one is the one I worry about most, because a lot of the people who taught me how to think about design came up through that exact tier of agency, and if it stops existing, I don’t know where the next cohort learns from.
I don’t have a clean ending for this one. The shape of the market is still in motion. I’ll come back to it later in the year when there’s more to say.
Jameson