Becoming a conversation designer in the AI era
What the job asks for in 2026 listings, how this course maps to those skills, and how to turn the capstone into a portfolio.
What the role is now
A conversation designer decides what an assistant says, when it asks, how it recovers and how it sounds. With language models, the job also includes the system prompt, the behavior of retrieval and tools, testing, and reading transcripts to find what to fix. Listings from 2026 ask for conversation flows and dialogue, prompt engineering, testing and QA, and experience with observability tools, with close work with UX, product and engineering. The role is not a copywriter for a bot, and not an ML engineer with a design deck.
Four skill areas
| Area | What you do | Where to learn it here |
|---|---|---|
| Dialogue and language | Persona as behavior, turn design, repair, handoff, inclusive writing | module 1, module 3, module 11 |
| Model and prompt behavior | System prompts, prompts per model, retrieval-grounded answers, context | module 5, module 6, module 23 |
| Testing and evaluation | Test cases, test sets, trajectories, judges, QA | module 14, module 15, module 16, module 24 |
| Data and product sense | Metrics, dashboards, safety, regulation, channels | module 17, module 18, module 12, module 13 |
Conversation design or prompt engineering?
Prompt engineering shapes what a model does. Conversation design shapes what the user experiences. In practice you do both. A prompt can be perfect and still produce a bad conversation if the turn design, the recovery and the handoff are wrong. This course teaches them together.
Build a portfolio from the capstone
The capstone (module 22) already produces what hiring managers want to see:
- a short transcript of the ideal conversation, written before any tool;
- the flow and where the assistant hands off;
- a system prompt and why each line exists;
- a test set with expected behavior and the checks that grade it;
- a metrics plan that says what "resolved" means;
- a risk note: what can go wrong, and the limit that prevents it.
Show your reasoning and your failures. A transcript that broke, why it broke and what you changed teaches a reviewer more than a polished demo.
Preparing for interviews
- Bring one conversation you improved and the evidence (before and after, with a test set).
- Be ready to explain a handoff rule and a guardrail you chose and the false positives it causes.
- Know how you would measure success without counting abandonment as containment (module 17).
- Be able to say what you would not automate.
A note on pay and demand
Many pages online quote salaries and growth percentages without a data source. This guide does not repeat them. Check listings in your own market, and use salary surveys that name their method.
Where to start
Follow the Conversation designer learning path on the course page, then do the capstone.