Algorithmic Recall

How Runway and Canva built the same feedback loop from opposite directions

On June 25, Runway launched Agent 2.0, and Canva launched Grow 2.0, two products that share a single architectural claim: performance data from published ads should feed automatically back into the generation of new ones. The two companies arrived at the same design from opposite starting points: Runway by adding marketing intelligence to a video generation pipeline, and Canva by adding AI creative generation to an existing platform for publishing and performance analytics. That two companies defined the problem in the same terms and built to the same destination, having started from opposite ends of the marketing tool stack, suggests that the gap between creating ads and measuring their performance has become concrete enough to attract solutions from multiple directions at once.

Unnecessary overhead

Running a digital ad campaign meant assembling at least three separate tools: a creative tool for building assets, a publishing platform for distributing them, and an analytics dashboard for measuring results. Each handoff between those tools required re-entering context, because the audience parameters, messaging direction, and creative rationale that informed each asset lived in the marketer’s head and never entered any shared system.

Each new campaign therefore began without any record of what prior campaigns had learned about which messages, formats, or audiences had driven results. Creation tools had no concept of campaigns, testing frameworks, personas, hooks, or messaging strategy, so each new session opened on the same blank canvas regardless of how much a brand had spent or learned.

The dedicated pipeline

Runway had already shipped Agent 1.0 on May 13, 2026, six weeks before Agent 2.0, as a conversational pipeline covering concept development, story beats, multi-scene generation, voiceover, music, and final assembly. That pipeline produced only video, with no internal framework for campaign structure, ad testing, audience personas, creative hooks, or messaging strategy.

Agent 2.0 layers marketing structure onto that production pipeline. Given a single prompt, the agent generates a marketing brief and creates campaign assets aligned to it. Users can upload ad metrics from Meta, YouTube, TikTok, or Google, and the agent surfaces insights from those results to guide creative revision, connecting past performance to future output. Finished assets pass through automated format adaptation, with the agent cutting platform-native dimensions for Reels, Stories, YouTube, and feed placements. The pipeline extends to localization as well, adapting copy and swapping visuals for each market without rebuilding the underlying assets.

Agent 2.0 draws exclusively on Runway’s own generative models, making every output contingent on a single generation system. The system automates planning, generation, and rough assembly but expects human oversight at three points: storyboard approval, regeneration of weak shots, and final quality review. Runway has described its next step as direct connectivity to ad platforms, with live performance data feeding new asset generation automatically, framing the finished asset as a starting point in a continuing cycle.

Oversight assembled

Canva built the analytics foundation for Grow 2.0 by acquiring MagicBrief in June 2025 for $22.5 million, a platform that had analyzed more than $6 billion in ad spend for brands including Fenty Beauty, Koala, and Linktree, and whose co-founder George Howes now leads Canva Grow. Four more acquisitions supplied the remaining layers of the platform: Ortto, a customer data and marketing automation tool with more than 11,000 customers; SimTheory, an AI agent management platform; Doohly, a digital out-of-home advertising platform; and MangoAI, an AI-powered creative intelligence tool.

Grow 2.0’s ad creation layer generates static and video ads in seconds by drawing on brand context, audience signals, and prior performance data. Magic Layers exports those ads into the Canva editor for refinement, and Bulk Publish then sends completed ads to Meta, TikTok, and LinkedIn in a single workflow. The Launch Dashboard provides a centralized view of campaigns across all three platforms. The performance loop closes through three tools working in sequence. Multi-Platform Ad Insights aggregates results across the connected platforms, and AI Ad Tagging labels each ad with structured data on what drove performance. Automatic Refresh then completes the cycle by generating fresh creative based on those tagged signals.

Grow 2.0 became available on June 25 in North America, Australia, and the United Kingdom, with additional markets to follow. Canva restricts AI Ad Tagging to its Business and Enterprise plans, placing the deepest performance-feedback capability behind the higher pricing tiers.

Convergent inspiration

Runway built Agent 2.0 from a creative foundation, treating video generation as the primary capability and building marketing intelligence on top of the production pipeline. Canva assembled Grow 2.0 from a workflow foundation, starting with publishing infrastructure and analytics and adding AI creative generation to a platform that already connected teams to Meta, TikTok, and LinkedIn. Runway’s practical advantage lies in creative generation quality, and Canva’s lies in publishing reach and a design workflow that teams already use.

All output from Agent 2.0 runs through Runway’s own Gen-4 system, which means that no individual scene can be routed to a different model, even when a specialized engine might handle that scene more effectively. Routing creation, publishing, and optimization through a single vendor provides workflow coherence but concentrates exposure to API policy changes from Meta, TikTok, or LinkedIn. If any of those platforms changes its access terms, teams whose entire campaign workflow runs through Canva have no fallback publishing path.

Runway’s loop currently extends from brief generation and asset creation to performance data analysis, while direct platform connectivity, which would allow live results to trigger new asset generation automatically, remains on the company’s published roadmap. Canva’s loop already spans the full cycle: AI creation, publishing across Meta, TikTok, and LinkedIn, performance reporting, and Automatic Refresh, which generates fresh creative from those results without requiring a manual brief. Both architectures aim at the same endpoint, a system in which performance data from a running campaign informs the generation of the next one without a human routing that information between tools.

Self-reinforcement

As Runway’s loop accumulates performance history, that history forms exclusively within what Gen-4 can generate, and each revision cycle inherits the model’s constraints as the only frame through which the system can measure improvement. Canva’s Automatic Refresh deepens platform dependency with each cycle. Fresh creative draws on aggregated results from Meta, TikTok, and LinkedIn, and the longer the loop runs, the more the system’s creative output depends on the continued availability of data from those three platforms. Across both systems, each iteration’s learning cycle passes through the same architectural constraints that the platform established at its foundation, and those constraints grow more binding with accumulated history.

A fully closed loop removes two human functions that had previously operated separately: carrying performance data between tools, and exercising judgment about which signals deserved a creative response. Generating creative directly from performance signals funnels each campaign toward the creative forms that measured well in previous ones, making past success the implicit brief for the next iteration. As both architectures mature, accumulated success narrows the range of what the system will propose, steering each cycle’s output toward patterns that prior cycles validated.