Will Craft Matter in the Age of AI Software?
We don't make things simply to have them exist in the world. We make things because the process of creation changes us.
Everyone can build software now. That’s the narrative.
AI coding agents write production code in minutes. Vibe coding lets non-technical people ship apps by describing what they want. The barriers to entry have collapsed. Software is commoditized.
So the big question now is: If anyone can build, does the “how it’s built” still matter?
The “SaaS is Dead” Chorus
You’ve probably seen the headlines. Satya Nadella declared “SaaS is dead” on a podcast in December 2024. The SaaS index dropped 6.5% in 2025 while the S&P 500 rose 17.6%. Klarna slashed 1,200 SaaS tools and halved their workforce. The narrative writes itself.
But what this narrative misses: SaaS isn’t dying. SaaS is evolving.
The seat-based pricing model is what’s dying. The static, “here’s your dashboard, figure it out” approach is what’s dying. The assumption that software waits for humans to do all the work is what’s dying.
Linear’s Karri Saarinen calls it “self-driving software.” Traditional business software facilitates human actions. It creates efficiency through constraints and nudges. But the human still does the work.
The next wave of software moves work forward on its own.
Think about this. Customer requests automatically extracted from support tickets and sales calls. New issues analyzed, researched, and assigned without human intervention. Simple changes dispatched to coding agents and completed while you sleep.
That’s not the death of SaaS. That’s the evolution of what software can be.
The “Build for Agents, Not Humans” Argument
Here’s where it gets interesting.
There’s a growing camp that says we should design software for AI agents as the primary users. “Agents are the new apps,” says HubSpot’s CTO. More than 80% of enterprises believe AI agents are replacing traditional apps as the system of work.
The argument is this: If agents will use our software more than humans, why optimize for human interfaces at all? Just expose APIs and let the agents handle everything.
I get the logic. And yes, we need to design for dual audiences now. Semantic HTML. Machine-readable schemas. Protocols like MCP and A2UI that let agents interact with our systems predictably.
But here’s what I think the “agents-first” crowd is missing: Agents still need humans to set intent, define outcomes, and make judgment calls.
Linear nails this with their autonomy spectrum. At one end, the system suggests actions like a lane-departure warning. You can ignore it. At moderate autonomy, the system takes the first pass and you correct what doesn’t fit. At full autonomy, you step out of the car at the end and confirm you arrived at the right destination.
Notice what’s consistent across all levels. Humans are still in control. Humans decide which projects to pursue. Humans define what “good” looks like.
Software isn’t becoming agent-only. It’s becoming human-directed, agent-executed.
And that distinction matters for craft.
Code Was Never the Moat
Here’s the uncomfortable truth the “software is commoditized” crowd keeps forgetting: code was never the moat.
It wasn’t the moat a decade ago. It’s not the moat now.
Distribution was the moat. Brand was the moat. Customer experience was the moat. The feeling you get when you use the product was the moat.
Yes, software was expensive to build. But expensive doesn’t mean defensible. You could always hire engineers. You could always outsource. The technical implementation was never what kept competitors away.
What kept them away was everything else. The understanding of user problems. The quality of the onboarding. The emotional connection to the brand. The reason someone chose you over the twelve other tools that technically did the same thing. That was the moat.
AI coding agents have made the implementation cheaper and faster. But they haven’t touched the hard parts.
They can’t tell you what to build. They can’t feel whether the interaction is delightful or frustrating. They can’t decide if a feature serves your users or just adds bloat.
If anything, the commoditization of code makes these things matter more.
The Photography Parallel
There are many parallels we can pull from the big shifts throughout humanity’s history.
But the most recent one that comes to mind is photography. Think about what happened with cameras.
Smartphones put a decent camera in everyone’s pocket. The technical barriers to taking a photograph disappeared. Anyone can shoot high-resolution images now.
Did that make professional photographers obsolete? Did the value of visual craft disappear?
The opposite happened.
Photography became ubiquitous, which meant the bar for standing out got higher. Everyone could take a photo, but not everyone could take a good photo. The difference between amateur and professional became more obvious, not less.
The same thing is happening with software.
Everyone can vibe-code a landing page or spin up a CRUD app. Which means there’s a lot more software in the world. Which means the software that actually stands out, that people actually choose to use when they have infinite alternatives, has to be better.
Not technically better. Experientially better.
The Slop Problem
Here’s the thing about AI-generated anything: we’re drowning in slop.
AI slop in content. AI slop in images. AI slop in code. The tools have gotten good enough to produce passable output at scale, and that’s exactly what’s happening.
The State of UX 2026 report from Nielsen Norman Group puts it bluntly: “Lazy AI features and AI slop are now ubiquitous, and the shine is fading fast. When everything gets an AI sparkle, it becomes noise, not novelty.”
The backlash is real. People are tired of interacting with obvious AI generation. They’re tired of chatbots that don’t actually help. They’re tired of features added just to check the “AI-powered” box.
What cuts through slop? Craft.
Products that feel intentional. Interfaces that anticipate your needs before you ask. Experiences that make you feel like someone actually thought about your specific situation instead of generating a generic response.
Elizabeth Goodspeed wrote an essay called “AI Can’t Give You Good Taste.” And she’s right. AI can produce technically competent output. It cannot produce output with genuine point of view, cultural understanding, or aesthetic judgment.
Those things still require humans. Those things require craft.
The Vibe Coding Paradox
I’ve been coding since I was 13. Sold my first software at 15, back in 2005. A buggy internet cafe management system for 50 euros that I had to maintain for months. I’ve experienced the evolution of software hands-on.
I’ve watched AI coding tools evolve from “cute party trick” to “genuinely useful” to “holy crap, it just built that in 20 minutes.”
And here’s what I’ve noticed: vibe coding (or agentic coding) reduces the implementation difficulty but shifts the burden somewhere else.
Scott Young captured this perfectly. He was building a flashcard app with AI assistance and realized: “It’s basically removed all of the actual implementation difficulty, but I’m still left with a lot of the conceptual difficulty of deciding what the behavior of the software should be.”
The AI didn’t suggest the design patterns. It didn’t propose the learning science principles to incorporate. It didn’t have opinions about user experience. It executed what he directed it to execute.
“If I didn’t nudge the conversation in those directions, the AI never suggested them spontaneously.”
That’s the paradox. And that’s exactly what I’ve been experiencing for the past 3 years building with AI.
A world where everyone can code is probably a world with a lot more slop. But it’s also a world with a higher bar for what counts as genuinely useful software.
The people who understand product, who have taste, who can make architectural decisions, who know what to build and why? They become more valuable, not less.
You were never paid to type code. You were paid to think. AI just gives you more space to do it. That is if you use it right.
What Craft Looks Like Now
So what does craft mean when AI handles the implementation?
It means product vision. Knowing what to build and, more importantly, what not to build.
It means design sensibility. Understanding the emotional impact of every interaction, be it user interface, chat prompts, or any other form of interacting with it. Feeling the difference between software that works and software that delights.
It means systems thinking. Seeing how pieces fit together. Anticipating edge cases. Building for the user’s real context, not the idealized happy path.
It means taste. The ability to look at AI-generated output and know “this is good” versus “this is passable” versus “this misses the point completely.”
It means the judgment to know when AI suggestions are wrong. Because they’re wrong a lot. And the people who can’t tell the difference are shipping slop.
The Product Builder Era
I keep saying we’ve entered the age of the product builder.
Not the software engineer who implements specs. Not the designer who makes wireframes. Not the product manager who writes tickets.
The product builder. The person who can do all of it, or at least understand all of it well enough to direct AI agents that do the implementation.
AI doesn’t replace this person. AI enables this person.
One person can now do what used to require a team. But only if that person has the judgment, taste, and product sense to direct the work effectively.
The barriers to building are lower than ever. Which means the differentiation has to come from somewhere else.
It comes from craft.
Software is Evolving, Not Dying
SaaS isn’t dead. It’s becoming agentic. Self-driving. Adaptive.
The interfaces are changing. We’re designing for humans and agents together. The pricing models are shifting from seats to outcomes.
But humans still need software. Humans still direct the work. Humans still define what “good” means.
And in a world flooded with AI-generated everything, the products that stand out will be the ones built with intention. Built with taste. Built with craft.
Everyone can take a photo now. But we still know the difference between a snapshot and art.
Everyone can build software now. The question is whether you’re building something worth using.
Craft isn’t dying. It’s becoming the only thing that matters.