Tylan “Chowder” Miller & Jeff Valin — The Survey Is Becoming an AI Agent | CYOG Episode 03

Surveys have a trust problem.

They ask people to flatten a complicated experience into boxes, rankings, and short answers. The business gets a data set. The person filling it out gets another form to finish.

That tension was at the center of this Create Your Own Gravity conversation with Tylan “Chowder” Miller and Jeff Valin. The two builders are looking at a different possibility: a survey that can listen, ask a useful follow-up, and help a business understand what it is actually trying to learn.

Who was in the room

Joe Moore and Katrena Drake hosted the conversation, with Tylan “Chowder” Miller and Jeff Valin in the speaker seats and members of the AI Mastermind contributing questions from the room.

Tylan brings the perspective of a developer, educator, and consultant who builds agents and prefers systems that can be owned rather than rented through a single vendor. Jeff brings the perspective of a marketer, automator, and operator who has spent years thinking about the difference between a form that collects information and a conversation that earns it.

The room kept returning to the same practical question: how do we make the experience more natural without pretending that a fluent agent is automatically a trustworthy one?

The question beneath the conversation

What is the survey actually for?

If the answer is “to collect responses,” the form may be enough. If the answer is “to make a decision, qualify a lead, understand a workflow, or build the right agent,” then the questions have to do more than fill a database. They have to help the next decision become clearer.

A survey is not the point

Tylan described a process in which a person’s answers can help shape the system that will serve them. Instead of asking an abstract set of questions and waiting for a later meeting, the information can be used to identify the platforms, goals, workflows, and constraints that matter in the actual business.

That makes the survey a kind of pre-onboarding conversation. It can clarify what the person is trying to automate, what they want to accomplish in the next 30 days, and whether the right answer is an agent, a workflow, a dashboard, or something much simpler.

Featured quote 1 — Build toward the conversation

The value is not automation for its own sake. It is the reduction of distance between what someone says they need and what they are able to try.

The conversation is the interface

Jeff pushed the conversation beyond the form. A survey can ask a person to choose from a list. An agent can notice an answer, ask a follow-up, and create a more natural path to the information the business needs.

That does not mean the agent should wander. The conversation still needs a purpose. It needs a clear boundary around what it is trying to understand, what it is allowed to ask, and what happens when the person wants a human.

Featured quote 2 — Make room for the human exchange

The difference between a helpful follow-up and an invasive one is not only the wording. It is the trust surrounding the question. People need to understand why the agent is asking, what will happen to the answer, and when the system will stop and hand the conversation to a person.

Bias belongs to the whole system

The group was asked who owns the bias when an AI agent probes for follow-ups: the model, the prompt, or the client brief.

The honest answer is that bias can enter through every layer. It can be in the question design, the assumptions about the customer, the data used to shape the agent, the rules around what counts as a useful answer, or the decision about which follow-up gets asked next.

That is why “make it conversational” cannot be the whole brief. The system needs boundaries. It needs an explicit purpose, restricted rules, and a human escalation path when the agent does not know enough to proceed.

A knowledge gap is not an agent failure

Jeff offered an important distinction: sometimes the agent is not failing. The knowledge is missing.

If a salon agent cannot answer the price of a haircut, the problem may not be that the agent is unintelligent. The business may not have given it the information. If the system does not know what a service costs, what the workflow allows, or which source is authoritative, a more enthusiastic prompt will not fix the underlying gap.

Good agent design includes the knowledge base, the tools, the rules, and the process for keeping all of them current.

Test the agent before it costs real money

Tylan described a build process that treats the first generated agent as a starting point, not a finished product. The agent still needs evaluations, failure cases, tool checks, knowledge checks, and testing across the situations a real user might bring to it.

Featured quote 3 — Production is not the first test

That is the difference between a demo and a dependable system. A demo proves that a path can work. Evaluation helps reveal how the system behaves when the path bends.

Human handoff is part of the design

An agent should not treat a request for a real person as a failure. Sometimes the best next step is to collect enough information to make the handoff useful. Sometimes the best next step is to stop asking questions and connect the person immediately.

The organization has to decide that rule before the agent is live. A human handoff is not an emergency patch. It is part of the experience, part of the trust model, and part of the promise the business makes to the person on the other side.

What the conversation makes possible

The survey is becoming an agent because the old form was never the real goal. The real goal was understanding.

AI can help a business ask better questions, notice patterns, qualify a need, and prepare a useful next step. But the system earns its place through restraint. It needs to know what it is trying to learn, what it is allowed to do, what it does not know, and when to bring a human into the room.

Better surveys are not about extracting more information from people. They are about creating enough clarity for both sides to move forward with less friction and more honesty.

Frequently asked questions about AI survey agents

What is an AI survey agent?

An AI survey agent uses a conversational interface to ask questions, respond to answers, and collect information in a more natural sequence than a fixed form.

Can an AI survey ask follow-up questions?

Yes, if the agent has a clear purpose, approved rules, and enough knowledge to understand when a follow-up is relevant. The follow-up should serve the decision, not simply extend the conversation.

Who is responsible for bias in an AI survey?

Bias can enter through the model, prompt, client brief, data, question design, and escalation rules. Responsible design examines the whole system rather than assigning responsibility to one layer.

When should an AI survey hand off to a human?

Whenever the person requests it, the system reaches a knowledge or authorization boundary, or the organization has decided that a human must make the next decision.

How should an AI survey be tested?

Test normal paths, edge cases, missing knowledge, tool failures, guardrails, handoffs, and the quality of the final output before putting the agent into production.

Enter the room

Before building the next agent, ask what decision the survey is meant to support. Then ask what the person on the other side needs in order to feel understood.

Hear the full Create Your Own Gravity conversation and keep the human exchange at the center. The best survey may not feel like a survey at all. It may feel like the first useful conversation.