I've spent the last two years building the AI backbone of a global product org, and I've come out the other side more optimistic about craft, not less. This is where I stand, and why.
Twenty years in design has taught me that every generation panics about the same thing: the tool that makes the craft faster will make the craft cheaper, and the craft cheaper will make it disposable. Photoshop didn't kill illustration. Figma didn't kill visual design. AI won't kill product design, but it will change, permanently, what's worth spending a senior designer's attention on.
At Monta I got to run that experiment for real: an AI backbone wired into every internal data source, a library of 250+ shared skills, a pipeline that let designers and engineers ship as peers, and a 30-agent system automating 90% of our product lifecycle. Weekly AI usage went from near-zero to 88% of the company in four months. Nobody was mandated to use it. They used it because it made them better at the parts of the job they actually cared about.
That's the sentence that stuck with me from that whole project. It's also the reason I don't think the interesting AI conversation is about adoption anymore. It's about judgment: knowing exactly which 10% of the work still needs a human being who has taste, context, and the willingness to be held accountable for a decision.
I stay close to a network of people building the frontier of this: folks at Anthropic, OpenAI, and elsewhere thinking hard about what responsible, sustainable AI-assisted creative work actually looks like. Not because I want to be first. Because I want the tools creative people are handed in five years to have been shaped by people who've actually done the craft.
The short version: AI is fundamentally transforming the product development lifecycle by compressing prototyping timelines, enabling synthetic user research, and automating code and design generation. That shifts human effort away from manual, pixel-pushing tasks toward deeper strategic problem-solving and systemic orchestration across teams.
What used to require a sprint of static mocks and a separate build phase to validate now happens in one sitting: high-fidelity, interactive, testable same-day. The bottleneck moves from "can we build it" to "do we know what to build."
AI-assisted synthetic research won't replace talking to real users, but it collapses the time between a hypothesis and a first directional read, so real research time gets spent on the questions that actually need a human in the room.
Design-system-aware generation means the handoff between design and engineering stops being a translation problem. The output isn't a spec anymore: it's a merge-ready PR, and the designer and engineer are reviewing the same thing.
None of this replaces strategic thinking. It makes room for it. The teams that win aren't the ones with the most AI tooling. They're the ones who used the time it freed up to get better at deciding what's worth building at all.
I think about a peer of mine whose design team survived a brutal layoff round by walking into the room with revenue and retention numbers instead of usage or adoption metrics. The same instinct applies here: the right way to justify AI investment isn't "look how much we use it," but the dollar figure it moved. That's the metric I actually track.
The best use of AI I've seen raises the floor for a junior designer and frees a senior one for the judgment calls only they can make. The worst use of AI treats craft as a cost center to be automated away. I build for the first version.
Tools trained on creative work owe something back to the people whose work trained them: in compensation, in attribution, or in access. I care where a tool's training data came from as much as I care what the tool can do.
Every rollout I've led shipped monitoring and guardrails alongside the tool, not after it. Speed without oversight isn't velocity: it's just risk you haven't noticed yet.
I've spent most of my career as a data lover: sponsoring analytics tools as an executive, fighting for instrumentation budget, being the person who actually reads the dashboards. So I understand better than most what AI-native analytics unlocks. Ask a plain-language question, get a real answer, no SQL, no waiting on a data team's backlog. That shift is genuinely thrilling, and I don't want to undersell how much better it makes decision-making.
But having been the executive sponsor for a dozen of these tools, my worry was never the model itself. It's the access model underneath it. The moment a natural-language layer sits on top of a data warehouse, every permission boundary your data team spent years building either holds or it doesn't, and suddenly anyone in the company can ask a question about a named user they had no business asking before.
That's why I pay attention to tools like Sundial: they're building the analyst layer with observability and governance as the actual product, not a compliance checkbox bolted on after launch. If AI is going to sit that close to sensitive user data, the tools sitting between the model and the warehouse need to earn that trust deliberately, not accidentally.
I'm not anti-access. I've spent my career trying to get more people closer to data, faster. I just think the teams that win this next wave will be the ones who treat governance as a feature people can feel, not a policy doc nobody reads.
I'm still writing this one. Check back soon.