Every week someone asks whether AI replaces the design cycle or just speeds it up. The honest answer is neither headline. Applied AI is useful at specific stages — mostly synthesis, iteration, and communication — and dangerous when you skip the stages that require judgment, context, and accountability.
This is how we think about it when we review portfolios live.
The cycle still wins
Whether you use double diamond language or something simpler, the work still moves through the same beats: understand the problem, explore options, make and test decisions, then communicate what happened and why it mattered.
AI does not remove any of those beats. It changes how fast you can move through some of them, and how polished the output looks before you have earned the thinking behind it.
Discovery and research: light assist, heavy human
AI is decent at summarizing what already exists — interview notes, support tickets, competitive scans, messy workshop output. It is not a substitute for talking to users, watching behavior, or sitting with ambiguity long enough to find the real constraint.
Use it to compress raw material. Do not use it to invent findings you did not earn.
Definition and prioritization: where judgment matters most
Framing the problem, choosing what to cut, deciding what trade-off you are willing to make — that is still designer work. An LLM will happily give you a plausible problem statement for whatever you feed it. Plausible is not the same as true.
If your case study jumps from a tidy problem slide to polished UI with nothing in between, reviewers notice. The missing middle is usually process: how you scoped, mapped, prioritized, and validated before you designed the solution.
Ideation and execution: faster surfaces, same bar
This is where applied AI shows up most in day-to-day practice — generating variations, rewriting microcopy, exploring layout directions, turning rough notes into structured outlines.
The bar does not drop. You still own the decision about what ships. Faster output just means you can explore more paths in the same amount of time, then apply taste and constraints the model does not have.
Communication: the underrated use case
The stage we reach for AI on most often during live reviews is not mockups. It is storytelling.
Recruiters and hiring managers skim case studies. They read headings, pause on one section, keep scrolling. If your section titles only say things like “Design requirements” or “Understanding the problem,” a skimmer learns almost nothing about your thinking.
One trick we use in reviews: rewrite each section title so it summarizes the paragraph beneath it — as if the title is the only thing someone reads. “The backend requires more discovery than users have time for” tells a story. “Design requirements” does not.
For iteration, we have been pointing people at Notebook LM — paste in a case study, generate a slide deck, and study the order it uses to tell the story. The images will be wrong. The narrative flow is often surprisingly useful. Steal structure, not screenshots.
That is applied AI in the communication phase: not replacing your authorship, but giving you a draft of how someone else might frame the same work.
Validation and impact: AI cannot fabricate proof
When you are targeting senior roles, reviewers look for evidence that the work moved something — revenue, behavior, efficiency, risk reduction. AI can help you wordsmith an impact section. It cannot create metrics you do not have access to.
When hard numbers are unavailable, we look for behavior change instead: what do users do differently because this shipped? “Analysts no longer export to Excel to share findings” is weaker than a conversion percentage, but it is real — and infinitely better than invented KPIs.
A practical filter
Before you reach for AI on any step, ask:
- Am I using this to think better, or to look done faster?
- Would I still stand behind this if a hiring manager asked me to explain it live?
- Does this output need my context, or is it generic enough that anyone could have generated it?
If the answer to the last question is “anyone could have generated it,” it probably should not be the most visible part of your portfolio.
Applied AI belongs in the toolkit. It does not belong in place of the cycle.