Practical AI notes.
Note: We used several AI models, each with custom standardized instructions and assistants. Prompts were tuned, versioned, and reviewed.
A few years ago, when AI hit the UX scene, it was either overhyped or dismissed. We approached it like any other tool: useful, not magical. These are a few notes on how AI helped us reduce delivery time by ~60%, from an estimated 240 hours to around 105. No headcount changes. No quality loss.
Here’s what worked, and what didn’t.
Note 1: Before the first meeting
Our SOP is to learn as much as we can before speaking with a client: history, news, products, leadership, and approach. It takes time. Now we use AI to gather and cluster sources: market signals, competitor moves, and user sentiment. In a few hours, we had a top-down view. Not perfect, but sharp enough to ask better questions. And better questions build better strategic relationships.
Takeaway: Use AI not to impress, but to prepare.
Note 2: The contract pass
We’ve read our contract templates dozens of times. Maybe too many. When we ran one through a legal-tuned AI model, it flagged vague language, edge-case clauses, and areas a client might challenge. Small issues, but the kind that send contracts back for revisions. This time, the client signed within a day.
Takeaway: Let AI reveal what routine makes invisible.
Note 3: Research synthesis
Pages of notes are slow to process. We fed AI our screenshots, transcripts, and notes. It surfaced contradictions, themes, and even accessibility flags. It didn’t decide. It just helped us see more clearly.
Takeaway: Use AI to spot signals. Your job is to confirm them.
Note 4: Visual exploration
The brief included custom illustrations. We asked AI to sketch early iconography. It gave us a solid start. Hours became minutes. Final polish was done manually. We don’t waste effort winning battles that don’t need to be fought. We let AI prepare vector sketches, so we can focus on making them production-ready.
Takeaway: If the first 70% comes fast, you have more energy to care about the last 30%.
Note 5: Code, with a caveat
Mid-project, the client proposed an idea that would raise the bar, but also increase dev time. Scope creep. First, we needed to test feasibility. AI agents helped draft the logic. Seven hours later, we had a working POC. But deployment stalled because security wasn’t aligned. Was it wasted work? Maybe. But it exposed friction we can now design for.
Takeaway: Speed exposes where alignment is missing. That’s still progress.
Note 6: The invisible work
Emails, meeting summaries, user stories. They are all necessary, all slow. AI drafted most of it. We still edited, but we never started from zero. It freed us to stay focused on the actual design.
Takeaway: The value isn’t in automation. It’s in reclaiming your focus.
The imperfections
AI stumbled. Visuals needed cleanup. The code needed debugging. Prompts evolved daily. But AI was never the final voice. We learnt how to communicate with AI to make it helpful, one small step at a time.
Takeaway: An assistant without ego is better than a peer with none of your context.
Some savings numbers
| Estimated | Actual | Saved | |
| Contract review | 5 days | 1 day | ~32 hrs |
| Research synthesis | 12–16 hrs | 1.5 hr | ~11–15 hrs |
| Visuals | 30+ hrs | 6 hrs | ~24+ hrs |
| Prototype dev | 32–40 hrs | 7 hrs | ~25–33 hrs |
| Admin & copywriting | ~24 hrs | 16 hrs | ~8 hrs |
Total reduction: ~135 hours (out of 240). Only existing platforms, used with intention.
Final note
“The version we presented today got compliments from multiple teams. Great reactions all around.”
That was enough.
One last takeaway
Use AI to get you thinking better than yesterday.
The article originally appeared on UX Design Lab.
Featured image courtesy: Mykhaylo Kopyt.
