How I actually use AI to build enterprise software in 2026 — and what models do what.
I've been designing enterprise dashboards and CRM systems for years. The complexity never changes: dense data, stakeholder chaos, and users who need answers in seconds.

What has changed is my stack. Here's exactly how I use AI today — not buzzwords, but the real workflow.

🧠 LLMs for reports & stakeholder communication
I use large language models (Claude, GPT-4o) to compress 40-page research reports into executive summaries. After user interviews, I feed transcripts in and get structured insight briefs — mapped to business goals. The LLM doesn't do the thinking. It does the scaffolding, so I can focus on the strategic layer. My synthesis still comes from me.

🔍 NLP for CRM intelligence
In enterprise CRM design, I use Natural Language Processing models to run sentiment analysis on support tickets, chat logs, and survey responses. Tools like AWS Comprehend and Azure Language Studio let me surface emotional friction at scale — before it shows up in churn data. Designing the CRM interface is one thing. Understanding what users are actually feeling when they use it is another.

📊 NLU/NLM for dashboard design
When designing enterprise dashboards, I use Natural Language Understanding models to map how users actually describe the data they need — versus what developers built. That gap is where bad UX lives. I run semantic clustering across user interviews with tools like Marvin AI, then use the patterns to redesign information hierarchies. The model finds the signal. I design the solution.

⚡ My actual research workflow in 2026:
→ Discovery: I use ChatGPT or Claude to rapidly orient myself in new domains (cloud architecture, logistics ops, fintech compliance) before writing a single research question. It's scaffolding, not truth.
→ Screeners & discussion guides: I use Copilot or Claude to turn assumption lists into first-draft discussion guides. Then I edit. Hard.
→ Thematic analysis: After interviews, I use Marvin AI for early thematic clustering. LLMs accelerate the first pass — but I own the actual interpretation.
→ Synthesis: I don't outsource this. AI can't read the room, notice hesitation, or catch the moment a user says "it's fine" while clearly frustrated. That's still mine.
→ Stakeholder reporting: I use LLMs to tailor the same insight for three different audiences — dev teams get friction maps, C-suite gets revenue implications.

Still figuring this out in public. If you're building enterprise products with AI in your workflow, drop a comment — I want to know what's actually working for your team.

#UXDesign #EnterpriseUX #ProductDesign #AITools #LLM #NLP #CRMDesign #DashboardDesign #UXResearch #HumanCenteredDesign