Runway unveils autonomous engine for ad campaigns

On September 30, Runway unveiled Runway Ads — an autonomous engine for performance marketing; the company's chief product officer Anthony Maggio presented the product on stage. The system connects to the client's ad accounts and brand kit, generates video and graphic ads itself, publishes approved variants to Meta, Google and TikTok, pulls performance data back from them, and on that basis assembles the next batch of creatives.

The company describes this as closing a loop that previously ran through three or four separate tools stitched together by hand: generation, publishing and measurement are brought into a single service.

The engine runs in a circle: made it, published it, checked the result, redid it

As input the system takes the brand kit, product footage, and the ads that have already shown the best results. It then produces variants and tags each one by hook type, audience, format and call to action — performance is then read at the level of these attributes, and separately for each market. Before a creative enters the approval queue, it is run through an automatic check for compliance with brand rules.

Among the stated capabilities:

• localization — the engine changes on-screen text, product screenshots and voiceover to fit the market, according to rules set by the client's team (for example, brand voice and terms that must not be translated);
• automatic re-layout for different aspect ratios and platforms' technical requirements;
• publishing to connected Meta, Google and TikTok accounts with consolidated statistics in one place;
• performance feedback that sets the next iteration of creatives.

For production teams this removes the most labor-intensive part of the work — making dozens of variants of a single clip for every platform, language and format. The company cites as an example a brand that, instead of a few key ads, gets dozens of hook and format variants in the same time it used to take to brief a single one.

How autonomous the engine is

Runway stresses that a human stays in the loop: by default, publishing requires manual approval, and switching to automatic mode per campaign or per creative type is a separate decision by the client. Autonomy is limited by budget rules: a maximum change in daily spend, a minimum share of new audience, and a cap on retargeting.

In the comments under the announcement, the word "autonomous" immediately raised questions — including whether the engine will spend the budget just as autonomously. So far the company's answer is a set of limits and an approval queue.

The numbers Runway shows on itself

The whole case is built on the company's own marketing. Since July 2026, Runway has been running its campaigns on the same infrastructure — primarily on Meta and TikTok — and reports that ad output grew from 77 to about 900 per week, return on ad spend doubled, conversion rose by about 34%, and the cost of acquiring a subscriber fell by 41% with click-through rate holding steady. The company attributes more than $100 million in additional annual recurring revenue to performance marketing.

Runway co-CEO Anastasis Germanidis explains the product's appearance by internal need: bringing generation, publishing and measurement into one system instead of three tools stitched together gave the company a full picture of paid advertising performance and allowed it to cycle through variants faster.

Runway does not disclose the methodology behind these measurements: absolute spend figures, subscriber numbers and campaign composition are not given, and all the metrics relate to the company's own program, not to client results. For now, Runway Ads operates in pilot mode with select enterprise partners, early-access requests are taken via a form on the product page, and pricing has not been announced.

Why Runway moved into advertising right now

The company calls Runway Ads a development of its work on Runway Agent, and in July it released Media Router, which distributes generation requests across image, video and audio models. Ads continues the same line — from the models themselves to products that organize their use — and adds budget spending on top of generation.

Against the general state of the market the move is understandable. Generative video has stopped being a rare technology: there are many models, they catch up with each other quickly, and selling access to generation is getting harder. Tying into the ad budget is a different economy: the client pays for the campaign's result, and the tool's value is measured in metrics the marketer already defends before management. For a company that raised $315 million in February at a $5.3 billion valuation, this is also a way to explain to investors where future revenue will come from.

The risk here is just as obvious: Meta and Google are developing their own AI advertising tools too — the very platforms Runway Ads publishes to and on which the performance data depends. A similar end-to-end scenario — creative generation, connecting ad accounts, launch and data-driven optimization — is already offered by other services, so independent players will have to prove they deliver creative quality and brand control that the platforms' built-in builders lack. For graphics specialists this probably means a shift in the work: less manual production of variants, more art direction, setting up brand rules, and checking what the engine puts on air; part of the routine contract work will likely shrink as a result.

Which of these would you trust a system with unsupervised?

• Assembling the clips
• Publishing to Meta, Google, TikTok