# How to test an AI phone agent in your browser before you buy a number

By Samuel Oram-Jones

You can test an AI phone agent on Bland by pressing **Talk in browser** and speaking into your computer's microphone, with no phone number, plan, or purchase involved. On Bland, your agent can talk as soon as you [create it](https://docs.bland.ai/build-your-first-agent), with no phone number involved, so you hear what you built before you buy a number.

## What do you need before you start?

You need a Bland account and an agent to talk to. [Building your first AI phone agent](/no-code/how-to/build-your-first-ai-phone-agent) covers creating one.

You also need a computer with a microphone for **Talk in browser**, or a phone number you can type in for **Call me**.

## How do you test an AI phone agent in your browser?

You test an AI phone agent in your browser on Bland from your **Let's get this agent talking** checklist, where a browser test is the second of four first-agent steps.

<!-- screen: Bland's dashboard showing the "Let's get this agent talking" checklist with Talk in browser and Call me, captured and dated for this page -->

1. Open your agent's own page. Once you have created an agent, your **Let's get this agent talking** checklist lives there rather than in Bland's dashboard.
2. Start the conversation. Bland gives you two ways, both on your checklist, and either one puts you in a real conversation with what you just built.
   - **Talk in browser** uses your computer's microphone. It asks for permission, then listens.
   - **Call me** rings a number you type into **Your phone number**.
3. Say something your agent should handle, and something it should not.

<!-- screen: Talk in browser during a test conversation with the microphone permission prompt, captured and dated for this page -->

## How do you know it worked?

You know your test worked on Bland when your agent answers you out loud and you can read the conversation back afterwards. [Building your first AI phone agent](/no-code/how-to/build-your-first-ai-phone-agent) covers promoting your agent to production and the conversation view that opens afterwards.

Three details matter on a first read. Pick an agent or caller message in a call's timeline in [Bland's Conversations](https://docs.bland.ai/agents/conversations) and the recording jumps to where that turn starts, even before you press play.

Hover a turn and the actions it supports appear; Raw LLM context shows exactly what your agent was given to work from. The right-hand column's Metadata section holds the conversation's ID, channel, direction, agent, start and end times, and a written summary.

## What do you do when your agent gives a wrong answer?

You re-run just that answer in [Bland's Testbed](https://docs.bland.ai/agents/testbed), which takes a single moment from a real conversation and runs it again with whatever you change. Testbed does not replay the call. It re-runs one turn, and the conversation after that turn is not re-derived.

1. Open the conversation you want to work from and hover the turn that went wrong.
2. Press the icon that appears, then pick what to test: **Dialogue**, **Route**, **Loop condition**, or **Variable extraction**. That choice is which decision from the turn you want to re-run.
3. Edit any of the messages in the middle pane: the instructions, the context, or what the caller supposedly said.
4. Set how many times to run the turn. Running once tells you what happens, and running several times tells you whether it happens reliably.
5. Read the result. Identical answers are grouped, each badged with how many of your runs produced it, and each marked **Matches original** or **Different from original**.
6. Press **Push change** to write your edited version back to your agent.

Five identical answers mean your agent behaves consistently, whether or not it behaves correctly, and five different answers mean the turn is a coin flip.

Since Testbed stops after one turn, a fix that works there still has to be tried against a whole conversation in Evaluations, Bland's test cases that run before you promote, once the Evals feature is enabled for your organization.

Norm, Bland's AI assistant, has direct access to your call logs and transcripts, so you can also point it at a problem, and [fixing a bad call with Norm](/no-code/how-to/fix-a-bad-call-with-norm) covers that route.

## What can go wrong?

Bland's Conversations list and Testbed have three common failures of their own. A blocked microphone and the Norma example agent answering in place of yours are covered in [building your first AI phone agent](/no-code/how-to/build-your-first-ai-phone-agent).

- **A conversation you know happened is not in the list.** Bland's Conversations list hides some conversations by default: Call source starts on Live calls, so simulated calls are out, a web chat appears only inside its agent, and older conversations may not be loaded yet. Change Call source, switch the selector at the top of Bland's sidebar to your agent, or press **Load more**.
- **I cannot find a button that says Open in Testbed.** On a conversation it is a test-tube icon with no label.
- **Push change is greyed out.** Your agent has changed since that conversation, or you opened Testbed from a test chat in Bland's agent builder. In both cases, copy your change and make it in Bland's agent builder.

[Testing before launch](/no-code/how-to/test-an-ai-phone-agent-before-launch), the [no-code AI call center](/no-code-ai-call-center) guide, and the [no-code AI phone agents](/no-code) hub cover the rest of a rollout.

Idaho Housing and Finance Association saves $750,000 a year by replacing its IVR with Bland, which handles 4,000 inbound calls a day and cut average call time by 20%.

> "We save thousands of dollars a day."

Greg Blake, Chief Information Officer, Idaho Housing and Finance Association

## Frequently asked questions

### Does Bland have a no-code option?

Yes. Norm, Bland's AI assistant, builds voice agents using natural language descriptions, and Bland's dashboard includes a visual builder for non-technical users.

### Do I need a phone number to test an AI phone agent on Bland?

No. On Bland, **Talk in browser** uses your computer's microphone, and nothing in Bland's first-agent steps needs a phone number, a plan, or a purchase.

### Can I test a Bland agent by phone instead of my browser?

Yes. On Bland, **Call me** rings a number you type into **Your phone number**, and it puts you in a real conversation with what you just built.

### Why can other people not call my Bland agent after I test it?

Your Bland agent most likely has not been promoted to production yet. [Building your first AI phone agent](/no-code/how-to/build-your-first-ai-phone-agent) covers that step, and a number comes after it.

### Does Bland's Testbed replay a whole call?

No. Testbed re-runs one turn from a real conversation, and the conversation after that turn is not re-derived, so you see what that one answer becomes when you change what led to it.

### Why does a Testbed re-run give a different answer when I changed nothing?

On Bland, the same request can produce different answers, which is why Testbed lets you set how many times to run a turn. Run it several times to see how wide the spread is before you treat one answer as your agent's behavior.
