Sixteen tools, four phases, three modes. This is the map for how we work alongside AI: what it runs, what it drafts, and what we keep for ourselves.
We don't ask "should AI do this?" once. We ask it sixteen times. The answer determines who's driving.
AI runs the task end-to-end. A human confirms the result before it goes anywhere.
AI drafts, summarizes, or proposes options. The writer keeps final judgment over taste, risk, and product fit.
AI sits this one out. The judgment, accountability, or taste isn't something we delegate.
Read it like a subway map. Lanes are AI modes; stations are tools; the x-axis is the phase of the work. Augment carries most of the line. Automate does the bookends, and Act alone shows up where taste matters.
Understand the problem space. Pull in PRDs, tickets, research, and competitors so the writing has the right scaffold before we name anything.
A buildup of context from PRDs, Jira tickets, and design notes that frames the problem for human and AI partners.
A structured review of peer/competitor UX content patterns to understand norms, opportunities, and differentiation.
Inventory and qualitative review of your own product content to find gaps, inconsistencies, and standardization opportunities.
A review of user language from research notes, clustering phrases into themes that surface real-world terminology.
Visually mapping concepts and their relationships to clarify domains and inform taxonomy decisions.
Define structure, terms, and intent. Draft the patterns that the strings will eventually fill. Every tool here is Augment: AI drafts; we decide.
A scripted prototype that simulates how the product talks to the user, testing comprehension, tone, and risk points before UI exists.
Defining the structure, attributes, and relationships of content types so content is consistent and reusable across surfaces.
A controlled vocabulary of preferred terms, definitions, and usage rules to reduce ambiguity and improve consistency across UI.
Wireframes annotated with intended messages, hierarchy, and tone, validating content strategy before final UI copy.
Draft, explore, and converge on copy. AI helps generate variants, but the convergence (what's right for the product, the brand, the legal team) stays human.
A structured document outlining the goals, requirements, and strategy for creating content for an experience.
Rapid exploration of copy options (variants, constraints, edge cases) to find the best strings and pattern-fit within components.
Collecting feedback (design, legal, brand, localization, accessibility) and iterating copy with traceability and implementation notes.
Ship, document, and govern. Automation does the verification heavy lifting; humans own the rationale that will outlast any one project.
A short, explicit record of why a content decision was made (tradeoffs, evidence, constraints) so future contributors can maintain consistency.
Making sure content in the experience matches what was designed, by comparing approved strings against implemented UI and flagging differences.
Specific, contextual instructions for translators and developers so content is accurately adapted for another market, culture, or language.
Ensuring written content is welcoming, respectful, and accessible, avoiding language that marginalizes, stereotypes, or excludes.
of our process is something AI either runs or drafts. But the moments that decide whether the work is good (concept mapping, critique, the final string call) are still ours.
Same content, three formats. One for skimming, one for sharing, one for sitting down with. If something is wrong, or the mode for a tool no longer feels right, tell the team.