Becoming a new dad forced me to start using AI design agents

I never expected fatherhood to change the way I design, but eight months in, it completely has. Somewhere between late night feedings and keeping three puppies alive, the long uninterrupted hours I used to rely on at my desk quietly disappeared. What replaced them was a phone in one hand and a real question about whether AI design agents could actually carry some of the load. This article is what I have learned using them as a new parent: where they genuinely help, where they still fall short, and where I think design and AI are heading next.

Key takeaways

• Becoming a parent cut my available design time to almost nothing, and that constraint is what finally pushed me to use AI design agents instead of just reading about them.
• I was late. By 2024, 59 percent of designers and developers were already using AI in their work, and by 2026 weekly use among designers had become near universal.
• "Agent" is still a loose term. In practice it means Claude plus a handful of design tools with agentic features, wired together by standards like Anthropic's Model Context Protocol, that let me explore directions while I am away from my desk.
• The tools I keep coming back to are paper.design, Magic Path, and Magic Pattern. Figma and Adobe have both shipped their own agents, though they still feel early to me.
• There is a real gap between a design agent and a designer. Figma's 2025 report found 78 percent of designers and developers say AI improves their efficiency, but only 54 percent of designers say it improves the quality of their work. That gap is the soul: color, typography, spacing, rhythm.
• The next frontier is context. Building visual libraries that tell an agent what good actually looks like, which is the problem I am building Hippo to solve.

Time disappeared, so I had to design smarter

Becoming a parent is a new experience in a lot of ways, so it is no surprise that time got scarce fast. I naively thought I would have at least one hand free while caring for my kid. Instead I found myself wishing Elon would hurry up and ship something that lets me control a computer with my mind.

I am not sure if I am being a little fortuitous here, using the article as an excuse to keep creating. But honestly I feel grateful for the whole experience, so why not write about it. Becoming a dad, taking care of a baby, and trying to keep my partner and me sane while we also have three puppies forced me to design smarter and lean on whatever tools were actually within reach.

That is where design agents, or design tools with agent features, started to matter. And I was late to it. By the time I caved, AI was already everywhere in our field. Figma's AI Design Trends 2024 report found that 59 percent of designers and developers were already using AI in their work. A year later it had gone from common to near universal: the 2026 State of AI in Design report found 91 percent of the designers it surveyed using AI at least weekly, up from 54 percent the year before. I had read about all of it. I just had not made it part of how I actually worked until I had no choice.

For me, the term agent has meant using Claude alongside a few tools with MCP capabilities to iterate on ideas and directions. In the small amount of time I do have as a new parent, that has helped me a lot. So I wanted to share where design agents are right now, where I am seeing real value, and where I think this is all heading.

What an AI design agent actually is right now

It is still early, and the word agent is doing a lot of work. The plumbing underneath it is newer than people realize. Anthropic only open-sourced the Model Context Protocol, the standard that lets AI assistants connect to the tools and data where you actually work, in November 2024. That standard is a big part of why a terminal like Claude Code can reach into a design tool and do something useful at all.

So when I say I use a design agent, I mean Claude in the terminal driving or assisting a design tool through that kind of connection, exploring variations on something while I am not sitting there. Really it is just a small stack of tools talking to each other, with Claude doing the driving. I am not alone in leaning on it either: in the 2026 State of AI in Design report, Claude was the most used model among designers, and 65 percent of them said they use Claude Code.

The tools that actually earned a place in my process

I use a lot of Claude in the terminal, so Claude Code is the starting point for almost any project lately. Anthropic seems to have a good grip on the market right now, since so many people begin with Claude or some instance of it.

The AI design agent market, though, is still very much up in the air. There are a lot of tools out there. I have tried most of them and found the ones that work for me. My short list includes paper.design, Magic Path, Magic Pattern, Wonder, Pencil, and more recently Figma's own agent feature. They are similar in how you work with an agent, but I have noticed a split: some lean fully into being an agentic design tool, while others let you work alongside an agent. Most of these apps are also very early. paper.design only raised its seed round, 4.2 million dollars led by Accel, in early 2025, and Magic Path launched to everyone for free in May 2025 on the back of a 6.6 million dollar seed. You cannot expect feature parity across tools this young.

Figma is probably the most longstanding tool here and is beloved by the creative industry, especially since you can now illustrate almost entirely inside Figma and skip something like Illustrator. paper.design has been really interesting to me and is one I keep open and frequent, in part because of the copy as React style handoff that turns a design into usable code. Pencil does a great job marketing the idea of design agents, giving you the ability to swarm on different directions at once. Then there is Magic Path and Magic Pattern. Magic Pattern is genuinely good and a little more flexible. Magic Path leans fully into Figma to clean code to production, so you can iterate on designs and actually ship them.

The big incumbents have noticed where this is going. Figma shipped its own Design Agent in 2026 for bulk edits, applying design systems, and filling layouts at scale, and Adobe used MAX 2025 to put an agentic AI assistant inside Photoshop. I have been in the Figma beta, and while it is fine, nothing about it has made me want to use only Figma yet.

The gap between a design agent and a designer

I am not here to say designers are cooked. But there is a real gap between a design agent and a designer right now, and that is okay. I am not in a rush to see that gap close and hand design off entirely.

The data backs up the feeling. In Figma's 2025 AI report, 78 percent of designers and developers said AI significantly enhances the efficiency of their work, but only 54 percent of designers said it actually improves the quality of that work, and only about a third said they can rely on its output. Designers have been more cautious about letting AI into the core craft too: just 31 percent use it for core design work, compared to 59 percent of developers using it for core development. A lot of the AI that designers do reach for is still text rather than visual work. The UX Tools Design Tools Survey found roughly 75 percent of AI use in design tools goes to text based tasks like copy and documentation, not the actual pixels. Even outside design, the 2025 Stack Overflow Developer Survey found 66 percent of developers frustrated by AI solutions that are close, but ultimately miss the mark. That phrase is exactly right.

When I review what an agent produces, some of it has moments of brilliance. By brilliance I mean it followed what I asked and explored real variations of it. But it can also be quite dumb, and that comes down to the foundations of design. I have said this for years: I probably capped out on my design expertise five or six years into my career. For me that meant mastering color, typography, spacing, layout, and rhythm, the foundations of what I believe makes a good designer. Design agents lack that right now. They lack the soul. The 2026 State of AI in Design report put the same idea more carefully, noting that AI may close the gap on craft and usability faster than on taste, judgment, and cultural foresight. Linear's CEO Karri Saarinen said it best in that report: "The hard part of design is rarely generating the form. It is understanding the problem well enough to know what and how something should exist at all." I am not betting against AI here. Right now, though, the work it produces on its own still feels infantile.

Designing with one hand

When you are low on time and you genuinely only have one hand free, that becomes a forcing mechanism. It pushes you to use these tools in a way that actually fits your process.

With an eight month old, I have spent the last couple of months designing with one hand, often without a keyboard or a mouse at all. I just use my phone whenever I can. The way I do this is with a remote desktop app called Jump that lets me sign into my Mac Mini from the room next door while I am feeding my kid. I can navigate, text my Claude terminal, and tell it to use a specific tool to iterate on something. I have started a task and then walked away to take care of my partner and my child without a second thought.

Sometimes I describe what I want from almost nothing, using speech to text to give a general sense of direction. Most of the time, though, I have already built something and I use an agent to iterate on variations, or to take on features and pieces I just cannot get to myself. This is where the real payoff shows up. Nielsen Norman Group's research found generative AI tools raised throughput by an average of 66 percent across three controlled studies, and Figma's 2025 report found that successful AI teams were far more likely to explore multiple directions at once, 60 percent of them versus 39 percent of less successful teams. Coming back to my desk to find a pile of iterations on the thing I have been thinking about all day is incredibly helpful. For me, that is what success looks like.

What this means for design teams

Because AI starts from code, it is easy to see that design agents will nail things like design systems, software, and websites. That part is clear.

There has always been this internal battle in the creative industry over fighting for a seat at the table, to be seen as more than a pair of hands executing ideas from sales, marketing, or the go to market team. AI agents are becoming that medium for a lot of people. Rather than burning a senior creative's time, a team can hand certain tasks to an agent.

What I have learned working with different startup teams over the last few months is that you need a foundation in place before you let a team run wild with agents and unsupervised design. A lot of teams still do not want to prompt a design agent. They do not want to switch context away from the thing in front of them just to get something out. So when the ask does come to design and creative, we end up leaning on these tools even harder and faster to keep feeding the rest of the business and help everyone move quicker. This is becoming the default. More than 80 percent of designers and developers in Figma's 2025 report said learning to work with AI will be essential to their future success.

If you compartmentalize the day to day tasks designers do, a lot of them will be handed off to an agent in some form. And I think that frees designers to think at a higher level and work on bigger problems. So AI design agents are very close to mainstream.

Context is the next frontier

If we keep compartmentalizing the creative process, the next area of real value is context. Building visual libraries that distill what you are actually looking for so an agent can act on it.

The money is clearly betting on this direction. Grand View Research estimates the broader generative AI market at about 16.9 billion dollars in 2024, growing to roughly 109.4 billion by 2030, a 37.6 percent compound annual growth rate. A lot of that spend is chasing exactly the gap I keep running into: tools are fast, but they do not yet know what good looks like for you specifically.

This is a shameless plug, but it is also exactly the problem I am building a tool called Hippo to solve. There are plenty of bookmarking apps that let you save things into a library where they mostly go to die. That has been my experience every time I have tried one. I wanted to build something with more utility and a bit more agency, so it actually does something with what you save. Hippo distills what you save down so AI can act on it or create things in the same spirit, similar to how image generation works today.

On the tools side, Magic Path has become interesting to me as a process: design in Figma, import into Magic Path, tune it slightly, then send it off to be developed and put into production by AI. Magic Pattern is similar with a little more flexibility. Both do a great job of distilling designs into clean, readable code. That matters, because Figma's own code output is still messy, which burns a lot of context and tokens and just feels slow. My favorite across the board is still paper.design. It is the most successful in its generations and output, and with the Quiver integration it becomes a real player. Figma's new agent is fine, but nothing about it made me want to use only Figma.

More time away from the keyboard, more present as a parent

The design landscape is shifting. Before having a child, and before AI and design really came together, I needed what I now think of as an extortionate number of consistent hours at a keyboard. Time to think, time to massage an idea, time to ruminate.

With AI design tools and agents, plus speech to text and Claude in the terminal, I get a little more time away from the keyboard. That freedom is what lets me raise my kid and still be present. For where I am in life right now, that is the whole point.

Frequently asked questions about AI design agents

What is an AI design agent?

An AI design agent is an AI model, often something like Claude, that can take a design instruction and act on it across one or more design tools, usually through a connection standard like the Model Context Protocol. It can explore variations, draft layouts, or turn designs into code while you handle something else.

Can AI design agents replace designers?

Not today. In the 2026 State of AI in Design report, around 80 percent of designers said they still rely on their own judgment over the AI's, and in Figma's 2025 report only 54 percent of designers felt AI improves the quality of their work. Agents are strong at speed, code, and iteration, but taste, typography, spacing, and judgment still rely on a human.

What are the best AI design tools right now?

It depends on your process. The ones I reach for most are paper.design for generation quality, Magic Path and Magic Pattern for turning designs into clean production code, and Claude Code as the agent that ties everything together. Figma and Adobe have both shipped agents of their own, though they still felt early to me.

How do designers actually use AI agents day to day?

The highest value use is iteration. You give the agent a starting point and let it explore multiple directions while you are away from your desk. Figma found that teams who explore more directions tend to succeed more often, and you can kick off that exploration from your phone if you have to.

Is it worth learning AI design tools as a designer?

Yes. More than 80 percent of designers and developers in Figma's 2025 report said learning to work with AI will be essential to future success, and weekly use among designers is now near universal. Even when the output needs a designer's hand, the time saved on early iteration is real.