AI agents
How I bring AI into the team
AI is not one person's shortcut. It is a method: where it goes, where it does not, and who reviews. That is how it scales in a large team without becoming a risk.
Where AI goes
In the steps where volume is high and the decision stays human.
- Research synthesis: transcripts, clustering and hypotheses in minutes
- Product writing: microcopy, test scripts, component docs
- Design system: tokens, variants and documentation generated and reviewed
- Working prototype: real code to test before spending engineering
- Data: reading funnels, queries and dashboard summaries
- Review: accessibility and consistency checklist before handoff
Where AI does not go
Guardrails I agree with the team before turning on any tool.
- Sensitive data: no customer, contract or proprietary code outside the approved environment
- Final decision: AI proposes, a person decides and signs
- Research with people: AI does not replace interviews; it prepares and synthesizes
- Craft: the quality bar is the same, with or without AI
How it becomes process
What I install so AI belongs to the team, not to one person.
- Playbook per step: prompt, input, expected output and who reviews
- Prompt library: versioned with the design system
- Human + AI pair: designer and dev reviewing together, same day
- Metric: time to first prototype and rework at handoff
- Training: short sessions, the team's real cases, no hype
How it becomes a result
Colbe: a SaaS from zero to live with AI at every step. Desenvolve AI hackathon: a working portal in one week and first place. This site: designed and coded with AI, reviewed by hand.
Where I used it
Research synthesis, copy, design system tokens, components and the comments server.
Where I did not
Positioning decisions, the case narratives and what goes live. That is mine.
How I reviewed
Every generated piece went through reading, browser testing and an accessibility audit.
The interview
An AI answers for me with the site content. Every conversation is saved and I read it later.
What I measure
Questions asked, interviews completed and what people most want to know.