Six months ago, I paid for Claude Pro with my own money. Nobody at the company asked me to. Nobody knew I was doing it.
I spent two weeks testing it on real work problems. Content generation. Strategic plans. Video scripts. When I had enough evidence that it worked, I walked into the CEO’s office and pitched it directly.
Today, we have 4 Claude agents running at Sell2Rent. Two people on the team use them daily. And the way I work as a product designer has fundamentally changed. Not in the way the industry articles describe, with talk of “agentic orchestration” and “ephemeral interfaces.” In a much simpler, more practical way.
This is what AI actually looks like in my daily work. No frameworks. No predictions about the future of design. Just what’s real right now.
What I Actually Use AI For (And What I Don’t)
I’m the sole product designer in a 35-person PropTech startup. I handle product design, CRO, marketing automation, SEO, and now AI tooling. That’s not a job description problem. It’s the reality of working in early-stage startups where adaptability matters more than any single skill.
Here’s where AI has genuinely changed my workflow:
Content production. The 4 Claude agents I built handle content generation, video scripts, strategic planning documents, and one more specialized task. Before these agents, producing a single piece of multi-channel content, LinkedIn, Instagram, YouTube scripts, blog posts, took me most of a day. Now the first draft takes minutes. The editing and refinement still takes time, but the starting point is radically better than a blank page.
CRM automation. When I migrated Sell2Rent from Zoho to HubSpot, I built the entire investor workflow from scratch: lifecycle stages, email sequences, automation triggers, matching logic between deals and investors. AI helped me think through the logic architecture faster. Not by building it for me, but by being a conversation partner for complex system design.
Vibe coding. I use tools like v0 to take Figma designs and generate functional front-end code. The back-end stays with the dev team. But the ability to prototype in real code instead of just high-fidelity mockups has collapsed the gap between design intent and implementation. That gap is where most design value gets lost.
SEO and analytics. I manage the full SEO stack for sell2rent.com using Ahrefs and Pagesense. AI helps with keyword analysis, content optimization, and pattern recognition across large datasets. Not replacing the strategic decisions, but accelerating the analysis that informs them.
What AI Still Can’t Do (The Final 20%)
There’s a number that keeps showing up in productivity research: AI accelerates typical tasks by roughly 80%. That matches my experience. But the remaining 20% is where all the value lives.
AI can generate a first draft of a LinkedIn post in my voice. It can’t tell whether the opening hook will make a founder in San Francisco stop scrolling. That’s pattern recognition built from years of writing, posting, and seeing what lands.
AI can run keyword analysis and suggest content opportunities. It can’t decide which opportunity aligns with the brand strategy I’m building for miguelba.com. That requires understanding the full picture of where I’m going, not just what the data says today.
AI can generate functional front-end code from a Figma design. It can’t tell you that the design itself is solving the wrong problem. I’ve shipped projects where the execution was technically flawless and the outcome was still a failure because the structural decisions around the design work were broken.
That’s the lesson from Keller Offers that applies directly to AI: tools don’t save you from bad process. A faster way to produce the wrong thing is still the wrong thing.
The 20% that AI can’t touch is judgment. Taste. The ability to look at something that technically works and know it’s not right. That’s not going away. If anything, it’s becoming more valuable because everything else is getting cheaper.
How I Actually Built the Claude Agents
I want to be specific about this because most articles about “AI in product design” are theoretical. This is what I actually did.
Step one was using my own money. I paid for Claude Pro before proposing anything to the company. This matters because it meant I could experiment without permission, without bureaucracy, without a committee deciding whether AI was “strategic enough” to invest in.
Step two was testing on real work. Not toy problems. Actual content that needed to get produced, actual strategic documents that needed to get written, actual video scripts that needed to match a specific voice and tone.
Step three was going directly to the CEO. Not to my manager. Not through a proposal document. I showed him what the agents could do with actual output from real tasks. He said yes.
Step four, the ongoing one, is refinement. The agents aren’t a finished product. They’re a process. I’m constantly adjusting the prompts, the workflows, the quality standards. Two team members use them daily now, Emilia and Jose, and their feedback shapes how the agents evolve.
The whole approach mirrors how I’ve always operated in startups: identify the problem, build a solution on your own time, prove it works with real results, then pitch it. The tool is AI. The method is the same one that got me from pure UI design to running CRO, marketing automation, and product strategy.
Vibe Coding: Where Design Meets Development (For Real)
The divide between design and development is collapsing. Not in theory. In practice. And it’s changing what it means to be a product designer. This connects directly to something I’ve written about before: the conversation between design and dev is where most product value gets created or destroyed.
At Sell2Rent, I use v0 to take Figma designs and generate functional React components. The back-end is entirely the dev team’s domain. But the front-end prototyping happens in real code now, not just in Figma. That means when I hand something to development, the conversation starts from a working prototype, not a static mockup that might or might not translate accurately.
This is significant because one of the most common failure modes in product design is the gap between what Figma shows and what the built product looks like. I’ve lived through that gap. At Keller Offers, the dev team adopted a different component library without consulting design, and the gap between prototype and product became permanent. The project failed partly because of it.
Vibe coding doesn’t eliminate that gap entirely. But it compresses it. When you can hand a developer working code alongside a Figma file, the conversation changes from “make it look like this” to “here’s a working version, let’s refine it together.”
The tools I’ve found most useful: v0 for Next.js components, Bolt.new and Lovable for full MVP prototypes when speed matters more than polish. Each has trade-offs. None of them replace a real development team. All of them make the designer-developer handoff less lossy.
The Honest Part: What Gets Harder With AI
Nobody talks about this, but AI has made some things harder.
Expectations inflate. When your team knows you can produce a first draft in 15 minutes instead of 4 hours, the definition of “done” shifts. More iterations. More variations. More channels. The speed increase doesn’t translate into less work. It translates into more output at the same pace. That’s a form of workload creep that’s easy to miss until you’re burned out.
Quality control gets more important, not less. AI output is confident. It reads well. It looks finished. But “looks finished” and “is finished” are different things. The designer’s job increasingly becomes the editor, the quality gate, the person who catches the 15% that’s subtly wrong. That requires more attention, not less.
The copyright and originality questions are real. When AI generates content or code, the provenance is unclear. For a personal brand built on authentic experience, this matters. Every piece of AI-assisted content I publish goes through a filter: does this sound like something I would actually say based on something I actually experienced? If the answer is no, it gets rewritten or cut.
What This Actually Means for Product Designers
The industry narrative is that AI is replacing designers. That’s wrong. AI is replacing the parts of design work that were never the valuable parts to begin with.
Generating a wireframe was never the hard part. Understanding what to wireframe was. Producing a design system was never the bottleneck. Knowing when to break the system for a specific user need was.
The designers who will struggle are the ones whose value proposition was speed of execution. If you were hired because you could produce Figma screens fast, AI does that now. The designers who will thrive are the ones whose value was always in judgment: knowing which problem to solve, knowing when the data is misleading, knowing that a technically correct solution can still be the wrong answer.
After 8 years in startups, from Mentive to Customela to Sell2Rent, the skill that has mattered most at every stage is the same one AI can’t replicate: the ability to sit in a room where nobody agrees on what to build and figure out the right next step. AI can help you execute that step faster. It can’t tell you which step to take.
The Real Shift
AI hasn’t replaced my job. It’s compressed the parts of my job that used to take the most time and freed me to spend more time on the parts that create the most value. The ratio has shifted from 70% production and 30% strategy to something closer to 40/60.
That’s the real story. Not agentic orchestration. Not ephemeral interfaces. Not eight pillars of anything. Just a product designer in a 35-person startup who paid for Claude Pro with his own money, built 4 agents, and now spends more time thinking about the right problems instead of producing deliverables.
The tools changed. The work that matters didn’t.
Want to see how I apply AI and design thinking in real projects? Check out my case studies or connect with me on LinkedIn.