# Where to go next

> Three ways to work with AQVEN, and where to connect a provider or read about how it works.

[Read in the browser](https://aqvenstudio.com/start/where-next/)

## Writing workflows in code

Define flows and nodes as files, write prompts and step code directly, and run everything from the command line. Start with [Engine](/engine/).

## Working in Studio

Build, run, and debug flows visually in the browser, with the graph, the run trace, and the checks all in one place. Start with [Studio](/studio/).

## Driving it through an MCP agent

Point a coding agent at your project over MCP and have it create, run, and fix flows for you. Start with [How to set up a coding agent outside Studio](/mcp-cli/set-up-an-agent-outside-studio/) — connect the server and paste one setup prompt — then the rest of [MCP & CLI](/mcp-cli/).

## Proving a flow works

A flow that answered one case right hasn't proved anything yet. [How to write an experiment](/engine/experiments/) puts one question in a file, [How to run a series](/engine/run-a-series/) answers it on many cases, and [How an agent takes a task to a reliable flow](/mcp-cli/research-loop/) is the whole loop, round by round, that a coding agent runs for you. [A day with AQVEN](/start/a-day-with-aqven/) shows that loop from the developer's chair.

## Connecting a model or a tool

Whichever way you work, a real flow eventually needs a model provider, an external MCP server, or a secret. See [Integrations](/integrations/).

## Understanding how it works

Read [the engineering loop](/concepts/engineering-loop/) to see how AQVEN takes you from an unexpected result to a verified fix, and [how a series decides](/concepts/how-a-series-decides/) for the statistics behind a verdict. If you're deciding whether to trust this with production load, start with [what this is built on](/concepts/what-this-is-built-on/) instead.
