Bottom line first: a 50-person design firm, 3 months of AI transformation, delivery time cut by 60%, and they only let go of 1 person.
Not 1 department. 1 person.
This blew up in industry WeChat groups — while "AI will replace designers" panic was spreading everywhere, this company went the opposite direction: barely touched the headcount, completely overhauled the workflow. I spent two weeks talking with their Creative Director, Chen, and dug through the entire process.
Background First
The firm is based in Hangzhou, doing brand design and spatial design, with a team of just over 50. About 30 are designers; the rest are strategists, project managers, business development, and admin.
Before 2026, their delivery model was traditional: strategist writes brief, designer produces 3-5 concepts, internal review, client review, revisions, final delivery. A mid-sized brand design project took 6-8 weeks from brief to delivery. Client fees ranged from 150K to 300K RMB.
The pain points were textbook: long cycles, endless revisions, severe designer overtime, stagnant profit margins. Chen told me their 2025 profit margin was 18% — not bad for the industry, but he felt it "wouldn't hold up much longer" because clients were getting more demanding, more cost-conscious, and competitors were undercutting prices.
3 Months, 3 Tools
In January 2026, Chen made the call to go all-in on AI transformation. He's not a tech guy, but he has a designer's instinct for new tools — you see something new, you want to try it.
Month 1: Midjourney V8.1 — Concept Art and Style Exploration
No surprise here. Midjourney V8 launched in March 2026, and V8.1 followed in April with native 2K rendering and 3x speed improvement. Chen got every designer an account and required that the concept phase of every project start with 50 Midjourney exploration images.
The effect was immediate. Previously, style exploration meant hand-sketching plus reference hunting plus moodboard assembly — at least 2 days. Now Midjourney produces 50 images in 30 minutes; designers pick directions and refine. This step alone was cut by 80%.
But Chen hit a snag: early on, a designer showed Midjourney-generated images directly to a client. The client loved them — until they noticed AI artifacts like wrong finger counts and garbled text. It was awkward. The new rule: AI output is internal reference only; anything client-facing must be hand-redrawn or deeply refined.
Month 2: Adobe Firefly — Brand Applications and Material Extension
Midjourney solved "creative divergence," but brand design isn't just concept art. After a logo is finalized, it needs to extend to business cards, envelopes, packaging, light boxes, VI manuals — that's "execution" work. Firefly's advantage is seamless integration with the Adobe suite: change a logo in Illustrator, and Firefly auto-updates all application mockups.
This step used to take 2-3 designers a full week. With Firefly, 1 designer finishes in 2 days. The freed-up capacity didn't go toward layoffs — it shifted to higher-value work like deeper brand strategy sessions with clients.
Month 3: ComfyUI — Custom Generation Pipelines
This was the big one. A young designer on the team with a tech interest spent two weeks learning ComfyUI and built a custom workflow: input brand color palette plus logo plus industry keywords, and it auto-generates a complete social media visual template set.
This used to take 3 designers two weeks per set. Now ComfyUI runs overnight, and the next morning designers arrive to 20 templates, pick 3 to refine, and deliver same-day.
But ComfyUI's learning curve is genuinely steep. Chen said out of 30 designers, only 4 actually learned it. The rest stick with the Midjourney plus Firefly combo. That's fine — not everyone needs to master every tool. The key is having someone who can build the pipeline; everyone else just follows it.
What Got Replaced, What Got Hired
The most sensitive topic.
The 1 person let go was a junior designer whose full-time job was "finding reference images." Frankly, this role has no reason to exist in the AI era — Midjourney's exploration output is faster, more accurate, and more diverse than any manual reference hunt.
But at the same time, Chen hired 2 new people: an AI tool engineer (the one who built the ComfyUI pipeline) and an AI prompt designer focused on writing and optimizing prompts. Net headcount: plus 1.
More importantly, the existing designers' work shifted fundamentally. Before, 70% of their time was "execution" — drawing, revising, layout. Now 70% is "judgment and decision-making" — which direction to pick, how to communicate with clients, how to control brand tone. Chen's words: "Designers finally have time to do what designers should be doing."
Workflow Redesign: From Assembly Line to Human-AI Collaboration
The old workflow was linear: brief to design to review to revision to delivery. Each person handled only their segment.
The new workflow is parallel:
- While the strategist receives the brief, AI has already generated 50 concept exploration images based on brief keywords
- Designers select directions from AI output while simultaneously using Firefly for brand application extensions
- ComfyUI runs social media templates and material extensions in the background
- Project managers see all output in real-time and sync progress with clients early
The serial workflow became parallel. That's the core reason for the 60% time compression. It's not that AI got faster — the workflow got faster.
The Numbers
By June 2026, 4 months into the transformation:
- Average project delivery cycle: 6-8 weeks down to 2.5-3 weeks (cut ~60%)
- Monthly projects: 8 up to 18
- Average project fee: roughly flat, but shorter cycles mean per-unit-time revenue more than doubled
- Profit margin: 18% up to 31%
- Per-designer output: up ~1.5x
- Overtime hours: down 70%
Chen emphasized one point: the biggest contributor to margin improvement wasn't "saving on labor costs" — it was "taking on more projects in the same time." Still 50 people, but capacity went from 8 projects to 18. Fixed costs unchanged, revenue doubled, margins naturally rose.
Pitfalls
It wasn't all smooth sailing. Chen shared a few lessons:
Pitfall 1: Copyright issues with AI-generated images. Some clients explicitly refuse AI-generated visuals due to unclear copyright ownership. Chen's solution: AI is internal reference only; final deliverables must contain human-created elements. This adds some work but avoids legal risk.
Pitfall 2: Designer resistance. Not everyone wanted to learn new tools, especially senior designers with 10+ years of experience. Chen didn't force it — he let the young ones go first, and once results showed, others followed naturally. "Don't argue with designers. Let them see peers using AI 5x faster — they'll get anxious on their own."
Pitfall 3: AI tool subscriptions aren't cheap. 30 Midjourney accounts plus Adobe Creative Cloud plus ComfyUI GPU servers — monthly tool costs approach 20K RMB. But compared to the labor and time savings, the ROI is clearly worth it.
3 Reusable Lessons
At the end, I asked Chen what advice he'd give other design firms looking to follow their path.
First: Don't start with "cutting headcount" — start with "boosting efficiency." AI tools should be introduced to help your existing team do more, not to employ fewer people. The moment you tie AI to layoffs, team resistance will kill any reform.
Second: Find the "tech believer" on your team and give them resources and time. Chen said without that young designer who taught himself ComfyUI, their transformation wouldn't have been this fast. Every team has someone like this — the question is whether you're willing to give them room to experiment.
Third: Workflow redesign matters more than tool selection. The same Midjourney — some firms use it for reference images, others for direct delivery. The gap is enormous. Tools are means; workflows are the end. Figure out where your workflow has bottlenecks first, then find tools to solve them — not the other way around.
Chen said something at the end that stuck with me: "AI won't replace designers, but designers who use AI will replace designers who don't." It's a cliché, but hearing it from a company that actually pulled off the transformation gives it a different weight.
Their plan for the second half of 2026 is to standardize this AI workflow into a SaaS product for small and mid-sized design firms. From "using AI to transform yourself" to "selling AI capabilities to others" — that might be the biggest opportunity in the design industry over the next 2 years.