What the job involves day to day
You build a bot that does one job well, such as answering questions, booking an appointment, or checking an order. Amazon Lex V2 is a good example of the classic setup. Its starter templates cover a customer support FAQ, appointment booking and order status. Each intent stands for something the user wants to do. You give it sample phrasings, the slots it has to fill (a date, a size, an account number), and a fulfillment step. AWS suggests writing that step as a Lambda function.
Google's Dialogflow CX models a conversation as a state machine. Pages are the states, and flows group related pages. The day-to-day work is wiring those states together, writing the prompts the bot sends when it fails to understand someone, and connecting webhooks to the systems that hold the real data, such as a CRM, a booking engine or a payments service. Most of the effort goes into edge cases.
Intents, LLMs, or both
Intent-based design gives you control and predictable behavior, but you have to anticipate how people phrase things. Large language models handle messy phrasing better, and the platforms now combine the two approaches. Amazon Lex has a feature called Assisted NLU that uses LLMs to improve intent classification and slot resolution. Google now builds agents in its Conversational Agents console with generative playbooks alongside flows, and the older Dialogflow CX console is marked deprecated.
In practice you decide which parts of a conversation must follow fixed rules and which can be generative. Identity checks, payments and anything with legal weight usually belong in fixed flows. Open questions answered from a knowledge base suit a model. Settings like banking and healthcare raise the stakes, because a wrong answer about a balance or a medication costs something real. Expect more review, logging and privacy work there.
Conversation design and testing
Google describes conversation design as a blend of voice interface design, interaction design, UX writing and related fields. In its framing, the designer maps what users can do and writes a detailed specification of the flow and its logic. On a small team the developer often does this work too. That means defining a persona, writing sample dialogs, then testing and iterating.
Testing is a large part of the job. Dialogflow CX supports test cases, experiments, and continuous tests tied to deployment. You run regression suites so a new intent doesn't break an old one, and you read real transcripts to find where people get stuck. Voice adds its own problems: speech recognition errors, interruptions, and prompts that have to be short enough to remember.
Handing the conversation to a person
A good assistant knows when to stop. In Microsoft Copilot Studio, a customer can ask for a live agent at any point. The bot then passes the full conversation history and the relevant variables to an engagement hub, which finds an available person. Google's Contact Center AI Platform shows escalation in the staff's CRM and lets you add a Skip to Human Agent button.
Google recommends putting at least one human agent on every queue that has a virtual agent, so people still have a way out if the bot fails. Building this handoff well, so nobody has to repeat themselves, is one of the most practical skills in the role.
Pay
USD a year, before tax. US figures from BLS OEWS May 2025, moved to 2026 with the Employment Cost Index. BLS has no conversational AI series, so these use software developers, the nearest match. That series covers the whole occupation, not chatbot specialists alone. State, employer size and seniority move the figure most. Refer to the Conversational AI & Chatbot Developers salary guide for figures from live adverts.
What you need
- ProgrammingYou need to write fulfillment code and webhooks, call APIs, and handle the data that comes back. Platform builders cover the dialog, but integrations still take real code.
- Dialog platform knowledgeKnow how intents, slots or parameters, flows and fallbacks work on at least one platform, such as Amazon Lex, Dialogflow CX or Copilot Studio.
- LLM judgmentDecide when a generative model should answer and when a fixed flow is safer. Then test both under realistic, messy input.
- Writing and designWrite clear prompts, sample dialogs and error messages. Short, plain wording matters even more in voice than in chat.
- Contact center integrationUnderstand queues, routing and handoff to a live agent with the conversation history attached.
- EducationBLS lists a bachelor's degree as the typical entry-level education for software developers as a whole. No license or required certification exists for this work.
How to get there
- StartLearn the software basicsGet comfortable with a programming language, REST APIs and JSON, since every bot calls other systems.
- First monthsBuild a task bot on one platformUse a template such as appointment booking or order status. Add your own intents and slots, then connect a real back end.
- NextAdd a generative layer and testsCombine fixed flows with LLM answers, write regression test cases, and read transcripts to find where users fail.
- Portfolio stageShip handoff to peopleWire escalation into a contact center queue and pass the conversation history along.
- First jobTake a first roleCustomer service, contact center software teams and in-house digital teams are common places for this work.
- Later careerSpecialize or move upGrow into voice, agent development or solutions architecture for contact center platforms.
Open now on AutomationRoles.ai
Questions people ask
Is a conversational AI developer the same as a conversation designer?
Not quite. Google describes the conversation designer as the person who maps the flow and logic in a detailed specification, then works with developers to build it. The developer builds and integrates it, though on small teams one person often does both.
Do chatbots still use intents now that LLMs exist?
Yes, many do. Amazon Lex still builds bots around intents and slots and uses LLMs to improve intent classification. Google offers generative playbooks alongside flows. Many platforms now let you mix fixed flows and generative answers.
How much does a conversational AI developer earn in the US?
BLS has no series for this role. The nearest match, software developers, has a 2026-adjusted US median of about $140,200, with a middle range of $108,500 to $177,400. That series covers all software developers, not chatbot specialists alone.
Do I need a certification or license?
No license exists for this work, and no certification is required. Platform certifications can help you show familiarity with one vendor's tools, but they won't get you hired on their own.
Where do conversational AI developers work?
Much of the work is in customer service: contact centers, the software vendors that serve them, and in-house teams that run support channels. Regulated sectors such as banking and healthcare also need this work, with more review and privacy controls.
Why does handoff to a human matter so much?
Bots fail, and stranded customers get angry. Google recommends at least one human agent on any queue with a virtual agent. Copilot Studio can pass the full conversation history so the person can pick up without starting over.
Sources
- AWS: Amazon Lex V2 core concepts
- Google Cloud: Dialogflow CX flow-based agent basics
- Google for Developers: What is conversation design
- Microsoft Learn: Copilot Studio hand off to a live agent
- Google Cloud: CCAI Platform, about virtual agents
- Google Cloud: CCAI Platform, assign virtual agents to queues
- U.S. Bureau of Labor Statistics: Occupational Outlook Handbook, software developers
