# Runway Agent Adds File Access and Code Execution: What Changed

Runway is extending its creative agent beyond generation, but the useful questions are about the work between assets.

By Paras Madan | 2026-10-31


Runway announced new Agent capabilities on October 30, including opening files and running code. The company says Agent can build brand launch kits, product shots, ad variations and motion graphics from references. The update expands the work its creative assistant can attempt, rather than introducing another video model.

The [official announcement](https://x.com/runwayml/status/2116227954518978729) directs users to the existing Agent product. It does not publish a separate API, a new model identifier or a benchmark for the added capabilities.

## What is new, and what was already there

Runway’s [release history](https://runway.com/changelog) matters here. Agent already supported campaign-oriented work, reusable Skills, Workflows and Brand Kits before this announcement. PDF input was documented in July, and creating or editing Brand Kits inside Agent arrived in September.

So the new claim should not be reduced to “Runway can read a brief” or “Runway can make an ad.” The explicit additions in this announcement are broader file opening and code execution. Those capabilities potentially give the agent more ways to handle the steps around generation, although the announcement does not enumerate supported file formats or execution limits.

That distinction also keeps this separate from Runway App Studio, announced earlier in the week. App Studio is framed around building and hosting a creative application; this update concerns what Agent can do while carrying out a creative task.

## Why code execution matters in a creative workflow

Generating a good image is only one part of producing a launch kit. A team also has to interpret references, maintain consistent dimensions, organize variations and decide which output belongs in which placement. Some of that work is easier to express as a repeatable operation than as another image prompt.

Code execution could help close those gaps. That is an implication of the capability, not confirmation that Runway has shipped every possible processing tool. The meaningful test is whether a repeat task becomes more dependable, with fewer manual handoffs, rather than whether the agent can produce a more elaborate demo.

For example, a team making several product-ad variants should still check that the product, offer and required text remain correct across outputs. **More automated steps create more places where a small mistake can spread.** A launch kit should be reviewed as a set, not approved because its first image looks right.

## Availability and the details still missing

Runway’s [Agent product page](https://runway.com/product/agent) describes Agent as available across its plans and explains that it can use different models for different tasks. That general availability statement should not be interpreted as a verified promise that every new capability has identical limits on every plan.

The October 30 announcement does not specify a code runtime, package installation policy, network access, file-size limits or a separate charge for execution. Those details matter before a team treats the feature as a production automation environment. They should be checked in the product and current account documentation rather than inferred from the phrase “run code.”

For now, the useful change to watch is how much of a real creative job stays in one conversation. If Agent can reliably carry references, intermediate files and repeatable operations through to reviewed deliverables, the benefit is less coordination work around generation. The quality of that whole path matters more than the number of tools the agent can call.

I’m Paras. I work at the intersection of AI, engineering and distribution.

I help technical founders turn launches and product capabilities into clear stories that developers and founders can understand and share.

[Book a 30-minute call](https://cal.com/paras-madan/30min).