An AI conversation designer defines how a chatbot, voice agent or AI assistant interprets requests, uses context, responds and takes action. The work now covers language, response patterns, structured content, tool rules, turn-taking, safety and evaluation.
That is the clearest signal in the current job market.
On 26 September 2026, Bot Jobs listed 301 live vacancies, including 28 in Conversation Design. Several newly listed roles are unusually explicit about what the profession is becoming.
T-Mobile describes its Principal Conversation Designer role as designing systems that shape AI behaviour, not writing interface copy or chatbot scripts. Deepgram wants a Staff Conversational Designer to work with machine-learning and engineering teams on turn-taking, barge-in, endpointing, latency and repair. Decagon combines conversation design with customer advice, evaluation frameworks and product feedback.
This week’s products help explain the change. Google’s new Gemini text-to-speech models add finer control over pace, emotion, dialect and backchanneling. SoundHound OASYS Edge can operate across cloud, edge and hybrid environments. NatWest’s voice-to-voice Spending Insights trial is designed for follow-up questions, topic changes and interruptions.
The conversation designer is no longer responsible only for what the agent says. The role increasingly shapes how the whole system behaves.
Key takeaways
- AI conversation designers create behavioural systems, not only scripts.
- Current roles ask for structured content, schemas, tool rules, evaluation and production learning.
- Voice AI adds turn-taking, interruption, latency, pacing and pronunciation to the design surface.
- Coding is not required for every role, but API, data and platform fluency are becoming important.
- A strong portfolio should show decisions, edge cases, actions and evidence, not only a polished happy path.
What does an AI conversation designer do?
An AI conversation designer turns user needs, business rules and technical capabilities into a coherent interaction. They decide what the agent should understand, how it should respond, which information it may use, what actions it may take and when it should ask, confirm, refuse or escalate.
Intents, sample dialogues, prompts, tone and repair still matter. The role has expanded because an AI agent may retrieve knowledge, maintain context, call an API, update a record and choose between several actions. The designer must connect the visible conversation with the system behind it.
Why is conversation design becoming systems design?
A generative agent does not follow one predictable flow. It interprets varied language and may choose how to pursue a goal. That flexibility is useful, but it creates decisions that somebody must make explicit.
Response patterns need structure
An AI agent needs reusable patterns for clarification, confirmation, refusal and recovery. These patterns govern behaviour across many journeys rather than supplying one line of copy at a time.
T-Mobile’s live Principal Conversation Designer vacancy calls for structured content models, schemas, ontologies and agentic interaction models. The person will help determine how agents interpret intent, negotiate outcomes and act across different levels of autonomy.
That is content design operating as product infrastructure.
Context must be designed
An agent needs to know what information is relevant now, what can persist and which source is authoritative. A designer may help define customer states, product entities, terminology, permissions and the conditions under which old context should be discarded.
This overlaps with information architecture and knowledge design. Poor context produces repetition, contradictions and actions based on stale assumptions.
Actions need boundaries
Answering a question and changing a customer’s account are not equivalent. The second action needs authority, validation and evidence.
Conversation designers increasingly define an action ladder:
- respond with general information;
- retrieve approved account information;
- suggest an action;
- ask for confirmation;
- complete an authorised action;
- decline or escalate when the boundary is reached.
Each step needs clear entry conditions and a recovery path. A fluent sentence cannot compensate for an unauthorised action.
Voice behaviour lives across the technical stack
Voice adds decisions that do not exist in text. When has the user finished? Is a brief “yes” confirmation or backchanneling? What happens when the user interrupts or the system is slow?
Deepgram’s live Staff Conversational Designer role owns persona, turn-taking, repair, confirmation and pacing. It also asks the designer to work with technical teams on endpointing, barge-in and latency budgets.
Google’s new TTS models extend that surface. Teams can direct delivery line by line, including pace, emotion, dialect and non-verbal cues. Greater control makes a voice and performance specification more important.
Quality must become repeatable
Conversation design used to produce flows for somebody else to test. Current roles increasingly ask designers to build rubrics, review production transcripts and turn failures into reusable evaluation cases.
Decagon’s live Conversation Designer vacancy spans chat, voice, SMS and email. It includes evaluation frameworks, playbooks, customer enablement and recommendations to product and engineering teams.
A strong designer can classify a failure, define acceptable behaviour and help the team verify the fix.
Which skills do AI conversation design jobs require?
The current vacancies repeatedly ask for six connected capabilities:
- Conversation craft: user goals, sample dialogue, turn structure, clarity, tone, progressive disclosure, repair and handoff.
- Behavioural architecture: reusable patterns for ambiguity, missing information, tool delay, failed authentication and requests beyond the agent’s authority.
- Structured content: taxonomies, entities, schemas, metadata, source ownership and rules for missing or conflicting evidence.
- Technical fluency: enough understanding of APIs, webhooks, authentication, retrieval and tool calls to design realistic behaviour and collaborate with engineers.
- Voice interaction: working knowledge of speech recognition, turn detection, synthesis, interruption, silence, names, numbers and latency-aware pacing.
- Evaluation and analytics: representative tests, expected actions and production evidence such as task completion, misunderstanding, escalation and repeat contact.
The live Conversational AI Designer role at Full Potential Solutions combines dialogue architecture with REST APIs, observability, incident reviews and customer-data controls. Our guide to Conversational AI evaluation covers the quality discipline in more depth.
Conversation behaviour may be distributed across prompts, platform configuration, speech services, business rules and code. Designers need to influence the people who own those layers and turn specialist decisions into one coherent experience.
What do current conversation design vacancies reveal?
These vacancies were live when checked on 26 September 2026. They show four distinct centres of gravity:
- Principal Conversation Designer at T-Mobile, Bellevue, Washington, US$115,400 to US$208,100: enterprise behavioural systems design across structured content, autonomy, explainability and compliance. Show response patterns, a content model and authority boundaries.
- Staff Conversational Designer at Deepgram, remote with Pacific-time alignment, US$180,000 to US$240,000: real-time voice design across turn-taking, repair, latency and evaluation. Show an annotated call or audio prototype with a supporting test set.
- Conversation Designer at Decagon, San Francisco or New York City, US$180,000 to US$220,000: customer advice, multilingual delivery, voice quality and reusable frameworks. Show a playbook or rubric that helps others apply your judgement.
- Conversational AI Designer at Full Potential Solutions, fully remote in the United States: voice and chat implementation covering platforms, APIs, analytics and incident learning. Show a production-readiness pack that connects design with monitoring.
Adjacent titles matter too. Sierra’s Strategist, Agent Development role owns an agent from scope through design, tools, evaluation, launch and improvement. Search by responsibility as well as title.
How do you build an AI conversation design portfolio?
Choose one narrow task, such as disputing a delivery charge, rescheduling an appointment or diagnosing a broadband fault. Build a behavioural-system case study with seven parts.
- Agent charter: Define the user, task, outcome and authority. State what the agent may answer, recommend and change.
- Conversation architecture: Map the routine journey, ambiguity, missing information, repair, refusal and escalation.
- Context and knowledge model: Identify important entities, customer states, approved sources and rules for conflicting evidence.
- Action specification: For each tool call, document the trigger, fields, permission, confirmation, success response and failure recovery.
- Voice behaviour specification: Define end-of-turn behaviour, interruptions, pace, pronunciation, backchanneling and unavoidable delay.
- Evaluation set: Write at least 20 cases covering routine requests, ambiguity, interruptions, missing data, unsafe actions and unavailable tools.
- Learning loop: Explain which production signals you would review, how a failure becomes a regression test and what would justify changing the design.
The Conversation Design Institute’s current Introduction to Agentic Experience Design course provides a structured learning option. Its published syllabus covers agent roles, context, knowledge, systems and trust across four hours and eight modules. Check the current access and subscription terms before enrolling.
How should you present conversation design experience on your CV?
Replace task descriptions with evidence of decisions and outcomes. For example:
- Designed reusable response and action patterns for ambiguity, confirmation, refusal and escalation.
- Built a structured content model linking customer intent, account state and approved knowledge.
- Specified voice-agent behaviour for turn-taking, interruption, repair and latency.
- Created an evaluation set and transcript-review loop that converted production failures into regression tests.
- Partnered with product, engineering and operations to connect conversational behaviour with APIs, permissions and service outcomes.
Use genuine measures where you have them. If the work is a personal project, label it as a portfolio case study rather than implying that it served real customers.
Frequently asked questions about AI conversation design careers
Is conversation design still a viable career?
Yes. Bot Jobs listed 28 Conversation Design vacancies on 26 September 2026. Employers increasingly expect work with agent behaviour, structured content, technical systems and evaluation.
Does an AI conversation designer need to code?
Not always. Design, language and systems thinking remain central. Basic fluency with APIs, data structures, retrieval and agent tools improves both your designs and technical collaboration.
What is the difference between a conversation designer and a prompt engineer?
A prompt engineer focuses on model instructions and context. A conversation designer owns the wider interaction: goals, turns, repair, content, actions, confirmation, escalation and quality. One person may cover both.
What should an AI conversation design portfolio include?
Include a user problem, agent authority, patterns, context model, tool behaviour, edge cases, evaluation and learning plan. Show how you make decisions, not only the final dialogue.
Can a UX writer or content designer move into conversation design?
Yes. Writing, information architecture and user-centred design are strong foundations. Add multi-turn interaction, agent behaviour, technical fluency, evaluation and evidence of how your work changes an outcome.
Conversation design did not disappear. It moved deeper into the product
The profession is not being reduced to prompt editing. It is expanding into the rules, structures and evidence that make AI behaviour understandable and useful.
The words still matter. So do the context model, action boundary, speech timing, failure path and feedback loop behind them.
Build one case study that makes those layers visible. Then explore current Conversation Design jobs on Bot Jobs, browse all Conversational AI opportunities and create a job-seeker profile so specialist employers can find you.
