AI gave designers superpowers, but they’re misused

Today he is Head of Product Design at ManyChat, the company behind comment-to-DM on Instagram and one of the fastest-growing companies serving the creator economy. In his first year he grew the design team from 8 to 21 people across two offices.

We got into why AI makes it easy to build the wrong thing fast, how a design team should split its time, why writing is now the designer's most useful skill, and the hiring workshop that starts with what would get someone fired.

Watch the episode on YouTube

Key takeaways#

What has AI changed about designing digital products?#

With AI in the room you can build anything extremely fast, which means you can also produce a lot of useless stuff. Alexander's rule: decide what to build before you start prompting. That includes deciding which decision points need a human. He's blunt about why the human stays: LLMs are pattern recognition machines, and people bring situational awareness. Going straight from "I heard this from a customer" to a solution gives you an AI-augmented feature factory, and your product idea can turn into "a very expensive hobby for your company."

Try this: before your next prototype, write down what you heard, why it's happening, and what it means in your context. Answer all three before anyone opens a prompt.

Is there still a place for the "Figma jockey"?#

Alexander doesn't think so. A Figma jockey, in his words, is a designer who ties their identity to a tool and a part of the craft that might simply go away. You can now take a good prompt or scenario, put it into Figma Make, and get a working prototype hooked up to your design system faster than anyone could draw it. The designers who adapt decide what to keep, what to throw away, and how to iterate. If you can't describe what to make, you create waste: tokens, mental load, attention and time.

Try this: take one screen you'd normally draw by hand and describe it in writing instead. See how far a prompt gets you, and where your judgement still has to step in.

How should a design team split its time?#

Alexander uses a rough model for software R&D teams like ManyChat and Mendix. Around 20 to 30% goes to evolving what's already shipped, because everything live should keep getting better. Roughly 50 to 60% goes to net new features, the things customers want tomorrow, or yesterday because a competitor already shipped them. The rest is reserved for future thinking: trying to leapfrog the competition, or disrupting yourself. AI collapsing delivery time is good news here, because it frees time for the fuzzy front end, where researchers and anthropologists turn weak signal into insight.

Try this: tag every item on your team's current roadmap as evolve, new, or future. If the future column is empty, you're not planning to leapfrog anyone.

How is AI changing design sprints and prototyping?#

Sprints get shorter, and testing gets earlier. With good input data and a clear sense of where you want to go, a five-day sprint can run in three, and you can test ideas after a couple of hours. At ManyChat, the design team uses AI tools mostly for prototyping and for co-creating with customers. Designers aren't pushing code to GitHub yet, and there's still a handover to developers. Alexander expects that handover to collapse, as it has at Intercom, where designers use Cursor within guardrails to ship front-end code. His warning: higher fidelity can be a distraction. The goal is getting the right product, and a polished prototype can make you feel you're there when you aren't.

Try this: cut your next design sprint from five days to three. Spend the saved days on the input research before it starts.

What is the most important skill for designers in the AI era?#

Writing. Alexander calls the ability to write and express your ideas clearly "the superpower" now, more useful to a designer than learning to code. The second half is better ideas. If you can build any idea, deciding which one matters, and de-risking it before you build, is where the work is. He warns against discarding what industrial and automotive design already know: they couldn't afford iteration, so they de-risked everything upfront, and it produced brilliant products. The trap for inexperienced teams is staying output-guided ("let's build this because we can") instead of asking what problem they're solving and what outcome they want.

Try this: ask every designer on your team to write a one-paragraph brief for their next piece of work. Review the writing before you review any screens.

Why do so many AI-built products look the same?#

Because the patterns are baked into the models. Alexander describes what comes out of every vibe-coding workshop: "a gray thing in the middle, a little bit of purple and a circular logo." LLMs are good at deductive work. They process more data than anyone on your team and surface patterns you overlooked. Novel ideas have to come from somewhere else, from people making inductive leaps and building an opinionated product. To stand out, you need intentional branding and positioning, and sometimes that means doing the opposite of what the model suggests.

Try this: put your last AI-generated prototype next to two competitors' products. If you can't tell whose is whose, start the branding conversation now.

What is the job of a design leader?#

To set a clear vision, assess reality honestly, and help the team and the functions around it ladder up to that vision. That means absorbing the tension between where you want to be and where you are. "Vision without action is just hallucination," so the vision has to be operationalised. Day to day, Alexander's first job is talent acquisition, which fills most of his calendar. The second is training the team on the inputs he needs to help them when they're stuck. And in a hyperscaling company, he pushes responsibility down relentlessly, until he's very uncomfortable with it. His proudest delivery proves the point: four years into Mendix, he went on sabbatical and the team, by then grown from 5 to 40, kept shipping. "Making yourself replaceable is the real achievement as a design leader."

Try this: pick one decision you made last week that someone on your team could have made. Hand it over next time, and tell them why.

What is the most underrated quality in a leader?#

Self-awareness. A design leader has to hold the tension between vision and reality without losing their mind or getting impatient, and that requires knowing what triggers you, what excites you, and what your values are. For Alexander it's a daily practice: breathwork when he wakes up, a cold shower, a few yoga asanas, then a moment sitting calmly to see what shows up. He calls that the bare minimum to stay performative under pressure. The responsibility is returning to baseline every day. His weekly martial arts class works on the same idea: drills build your capacity, so you stay calm "when shit hits the fan."

Try this: block 10 minutes at the start of tomorrow with no screen. Notice what's on your mind before the calendar decides for you.

How do you write a job description that attracts top design talent?#

Write it with intent, from the reality on the ground. Alexander hires for capabilities and blind spots, working from a roadmap of what the team needs in six months and a year. He's on the fifth iteration of his hiring framework, built over seven years. His practical tip is never starting from a boilerplate or another leader's job description. Instead he runs a performance profile workshop, an idea from Liftoff!, Chris Avore's book on design leadership. The team answers: what would get this person fired, what would make them quit, what will they do in the first year, and what will get them promoted? At ManyChat he's using it to hire a data designer, a skill no one on the team has yet.

Try this: run the four questions with your team before you write your next job post. Start with "what would get this person fired."

How do you give a design team space to learn new AI tools?#

Build it into the rhythm, and watch for burnout. Alexander admits AI makes work more intense, and prompting is addictive, so knowing your limits matters. ManyChat's design team has access to ChatGPT Pro, Claude Code, Cursor, Notion and Gemini, and he encourages every designer to download Cursor and start building. Two-week sprints now include time set aside to explore and learn. Weekly rituals and Slack threads let people share how they're prompting. You can't demand that someone grows, so he ties learning to each person's purpose through growth conversations. He uses Kim Scott's three conversations: life story, dream story, then a capability story. His own mental shift: "You don't need to know everything. You just need to know what to ask."

Try this: start a weekly 20-minute slot where one designer shows exactly how they used an AI tool that week. Keep it practical, and keep it short.

How do you keep trust when AI talks on someone's behalf?#

You let the culture decide how much automation shows. ManyChat is the company behind comment-to-DM: a creator asks followers to comment a word, and an automation starts a one-to-one Instagram conversation. The hard part is trusting a system to represent you in the right tone, without drift. Alexander sees big cultural differences. Creators in LA work in a hustle culture and are extremely commercial. In Germany, people are more hesitant, don't give away their data, and creators would rather flag a message as automated than risk breaking trust. Without automation, a creator who goes viral can spend two hours a day answering DMs.

Try this: if your product speaks for your customers, test one market with a clear "this is automated" label and compare how people respond.

---

Listen & watch#

• Watch the episode on YouTube
• Listen on Spotify
• Listen on Apple Podcasts
• All episodes at infinum.com/delivered DELIVERED is a production of Infinum and Your Majesty.