AI is reshaping how design agencies work. Here is how Feelpixel uses AI across real projects, what it actually saves, what it does not replace, and why skeptical intelligence is the most important skill a designer can have right now.
Everyone is talking about AI in design. We want to talk about what it actually looks like on a Monday morning.
Not the pitch deck version. Not the conference keynote version where every problem is solved by a single prompt and the output is perfect on the first try.
The real version. Where AI is genuinely useful some of the time, genuinely frustrating some of the time, and where the designer still has to make every decision that actually matters.
At Feelpixel we have been working with AI tools across research, ideation, visual exploration, and content structuring for a while now. And the honest answer to “does it save time?” is: it depends on what you are trying to do and whether you know how to question what it gives you back.
That second part is what we call skeptical intelligence. And it is the thing no tool can give you.
First, what does AI actually do in a design process?
It is worth being specific here because the conversation around AI and design tends to oscillate between two extremes. Either AI is going to replace designers entirely or it is just a fancy autocomplete. Neither is accurate.
What AI tools actually do well in a design context is compress the time between starting and having something to react to. That is genuinely valuable. A blank page is one of the most expensive things in a creative process. When AI can give you five directions to respond to instead of zero, the work starts moving faster.
According to McKinsey’s 2024 State of AI report, organisations that have integrated AI into creative and design workflows report a 30 to 40% reduction in time spent on early-stage ideation and content drafting. That tracks with our own experience. The early phase, where you are just trying to get the first rough shape of something, is where AI contributes most.
But the moment the work requires judgment, context, user understanding, or nuance, the tool steps back and the designer steps forward. That handoff is the thing most conversations about AI in design miss entirely.
What skeptical intelligence means and why it matters
Here is a term we use a lot internally at Feelpixel: skeptical intelligence.
It does not mean being suspicious of AI or refusing to use it. It means using it with your eyes open. It means knowing that AI is trained on patterns and will always give you the most pattern-consistent answer, which is often the most average answer. And in design, average is rarely what you are looking for.
When you ask an AI tool to generate a user flow for a fintech onboarding experience, it will give you something that looks like every fintech onboarding experience that has ever been documented on the internet. It is coherent. It is familiar. And it is probably wrong for your specific user, your specific context, and your specific product.
Skeptical intelligence is the ability to look at that output and ask: what does this get right, what does it miss, and what does it reveal about assumptions I was making that I had not examined yet? That is a designer’s skill. It is not something the tool can do for itself.
The designers on our team who use AI most effectively are not the ones who trust it most. They are the ones who interrogate it most. They use the output as a starting point for their own thinking, not as a destination.
How we actually use AI at Feelpixel
Research and synthesis
User research generates a lot of material fast. Interview transcripts, survey responses, usability session notes. Synthesising all of that into patterns used to take days. AI has meaningfully compressed that.
We use AI to do a first pass on transcript analysis, pulling out recurring themes, notable quotes, and apparent contradictions. It does not replace a designer reading the research carefully. But it means the designer starts that careful reading with a map rather than a blank page.
When we worked on Socleus, a cybersecurity platform for managing security events and threats, the user base was highly technical. The research conversations were dense and specific. AI helped us organise and categorise early findings so the team could spend their time interpreting the data rather than just sorting it.
Ideation and visual direction
Early in a project, when we are exploring visual directions, AI image generation tools let us move through a much wider range of aesthetic territories than moodboarding alone would allow. We can react to ten visual directions in the time it used to take to find reference images for three.
But this is also where skeptical intelligence matters most. AI-generated visual references tend to cluster around what is already trending. Left unchallenged, they push design work toward the familiar rather than toward what is right for the brand and the user.
When we worked on Gist Impact, an ESG investor portal, the visual language needed to feel credible to a sophisticated financial audience while also communicating environmental purposes in a way that did not feel like greenwashing. No AI tool was going to navigate that tension without a designer steering it deliberately. We used AI to generate early visual explorations and then moved away from almost all of it as the brand direction became clearer.
Copywriting and microcopy
This is probably where AI saves us the most time in a straightforward way. Writing placeholder copy, error messages, empty state text, onboarding instructions, and tooltip content is genuinely time-consuming and easy to deprioritise. AI handles first drafts of this material quickly and consistently.
On Orbit, a car management app that uses AI-driven insights to simplify ownership, we used AI to draft a large volume of contextual microcopy across different car states, notifications, and alert types. A human writer reviewed and refined everything. But the starting volume that AI produced in hours would have taken days to generate from scratch.
Design documentation and handoff
Writing design rationale, component annotations, and handoff notes is important work that often gets squeezed at the end of a project. AI has become a useful tool for drafting this documentation from design notes and conversations, so that the handoff to development is more thorough than it might otherwise be.
When we delivered Levo, a cybersecurity API security product redesign, the complexity of the interaction patterns required detailed documentation for the development team. AI helped structure and draft that documentation from our design notes, saving several days of writing time at the end of an already intensive project.
Is it saving time or adding complexity?
Both, honestly. And that is a more useful answer than most agencies will give you.
The time savings are real but they are concentrated in specific parts of the process: research synthesis, early ideation, first-draft copywriting, and documentation. In those areas, AI has meaningfully reduced the hours required and in some cases has changed what is possible within a given project timeline.
But AI also adds a new kind of work that did not exist before. Someone has to evaluate every output. Someone has to decide what to keep, what to discard, and what the output is revealing about a problem that the team had not yet articulated. That is not a small task. On projects where the team is not disciplined about this, AI can actually slow things down by generating volume that still needs to be processed carefully.
The net result, in our experience, is that AI makes good designers faster. It does not make average work good. And it absolutely does not replace the judgment that comes from understanding users, understanding context, and understanding what a product is actually trying to do in someone’s life.
Where we draw a clear line
There are parts of the design process where we do not use AI and are deliberate about that.
We do not use AI to generate user personas or journey maps from secondary research. These artefacts need to be grounded in real conversations with real users. An AI-generated persona is a statistical average dressed up as a human being. It will lead the design in the wrong direction every time.
We do not use AI to make final visual design decisions. It can generate options. A designer chooses, and the choice has to be grounded in brand understanding, user context, and design intent that the tool does not have access to.
And we do not use AI to replace the moments of genuine creative thinking that make the difference between a product that works and a product that resonates. When we worked on Intelligent Surveillance, a next-generation video management system for complex monitoring environments, the core design challenge was making high-stakes, real-time information feel manageable rather than overwhelming for operators under pressure. That is a human problem that requires human understanding to solve. AI helped us move faster in the early stages. It did not help us solve the actual problem.
What this means for the future of design agencies
Here is what we believe at Feelpixel. AI is not a replacement for design thinking. It is a pressure test for it.
If a designer’s value was primarily in producing visual output, that value is genuinely under pressure from AI tools. But if a designer’s value is in understanding people, framing problems correctly, making judgment calls in conditions of ambiguity, and knowing when the obvious answer is wrong, that value is not under pressure at all. It is more in demand than ever, because now you need those skills to evaluate what AI produces as well as to create the work yourself.
The agencies and designers that will thrive are the ones that treat AI as a fast, tireless collaborator with no judgment and use their own judgment to direct it well. Not the ones that either ignore it or hand over their thinking to it.
Skeptical intelligence, knowing how to use a powerful tool without being used by it, is the design skill of the next decade.
A note from the designer
Pooja Anant Kulkarni, Product Designer at Feelpixel
I will be honest. When AI tools started becoming genuinely capable, my first reaction was not excitement. It was a quiet kind of unease. Not because I thought it would take my job, but because I was not sure yet what my job would look like with it in the room.
What I have figured out over time is that AI is a very fast, very confident collaborator that has no idea what it does not know. And that combination, speed plus confidence minus self-awareness, is actually dangerous if you are not paying attention. It will give you a polished answer to the wrong question and you will not notice unless you are looking carefully.
The projects that have gone well with AI involved are the ones where we stayed genuinely curious about what the tool was missing, not just grateful for what it produced. On Tata AIG, on Risk Profiler, the AI gave us useful starting points. But the decisions that shaped those products, the ones that made them right for the people using them, were made by people who understood those users. Not by the tool.
I think skeptical intelligence is just another way of saying: keep thinking. Do not outsource your judgment to something that does not have any. Use the speed. Question the output. And never confuse a fast answer with a good one.
Pooja Anant Kulkarni is a Product Designer at Feelpixel with 5+ years of experience crafting purposeful digital experiences. She writes about UX strategy, product design, and the thinking behind interfaces that actually work.
Frequently Asked Questions
Is AI replacing product designers?
Not the ones who know how to think. AI is replacing the most mechanical parts of design work, first-draft copy, basic layout exploration, research sorting. But the parts that require genuine understanding of users and context, framing the right problem, making judgment calls, knowing when the obvious solution is wrong, require a human designer. If anything, those skills are more valuable now because someone has to direct and evaluate what AI produces.
What AI tools does a design agency actually use?
It varies by phase and task. For visual exploration, tools like Midjourney and Adobe Firefly. For research synthesis and writing, tools built on large language models. For rapid prototyping and layout exploration, AI-assisted features inside Figma and similar tools. The tools matter less than the discipline around how they are used and how the outputs are evaluated.
What is skeptical intelligence in the context of design?
It is the ability to use AI output as a starting point rather than a destination. To ask what the tool got right, what it missed, and what its answer reveals about assumptions you had not yet examined. It is not about being suspicious of AI. It is about staying in charge of your own thinking while using a very fast, very capable tool that has no judgment of its own.
Does using AI mean the design work is less original?
Only if you let it. AI generates pattern-consistent output, which means it tends toward the familiar. The designer’s job is to use that output as raw material and push past the familiar toward something that is genuinely right for the specific product, brand, and user. The risk of generic output is real. The answer is not to avoid AI but to bring more judgment to how you use it.
How do you know when to use AI and when not to?
We use it when the task benefits from speed and volume: early exploration, first drafts, research organisation. We step back from it when the task requires genuine user understanding, brand judgment, or creative decisions that need to be grounded in context the tool does not have. The line is roughly: AI for starting, humans for deciding.
Work with Feelpixel
Design is changing fast. The agencies that navigate this well are the ones that use new tools without losing the thinking that makes design work in the first place.
Feelpixel brings together UX strategy, product design, and a clear-eyed approach to how AI can genuinely improve the work without replacing the judgment behind it.
If you are building a product that needs both, let us talk.
