Here’s what’s on our plate today:

• 🧪 Why AI is eating the software workflow.

• 📰 Waymo recalls 4,000 robotaxis, anime meets the keyboard, AI compresses project timelines.

• 💡 Roko's Pro Tip: ask what outcomes your software owns, not what tasks it speeds up.

• 🗳️ Poll: What does AI eating the workflow mean for software?

Let’s dive in. No floaties needed…

The Laboratory

TL;DR

• The workflow is the product: Runway Agent turns a single prompt into a finished video, handling scriptwriting, generation, editing, and sound in a single conversation, collapsing the entire production pipeline.

• The Japan bet: Runway opened a Tokyo office and committed $40M, targeting the anime, ad, and gaming industries, built on structured pipelines that agents can absorb.

• Not an outlier: Coding agents like Devin and enterprise automation show the same arc, assistance hardening into autonomous execution that owns the entire task.

• Pricing breaks next: IDC expects most vendors to be off pure seat-based pricing by 2028. If one agent does five people's work, charging per seat undercounts the value.

• What's at stake: Owning the embedded workflow captures durable value, but Runway ($5.3B) faces Google and OpenAI with far deeper compute.

Why AI is eating the software workflow

For most of the last three decades, the software industry's business model has been built around helping people work faster. A brief became a draft, the draft underwent edits and revisions, and the final product emerged from a series of human decisions supported by digital tools. The software improved the process, but the human remained at the center of the work. For most software companies, that arrangement was the business.

That arrangement, however, is beginning to change in ways that are far more structural than incremental. Across areas such as video creation, software coding, voice generation, customer support, and office operations, AI systems are doing more than simply assisting people at each step of a task. In many cases, they can now take a single prompt and produce something close to a finished product on their own, reducing the need for many of the intermediate steps that once required human involvement. As this happens, the workflow itself, long treated as the foundation of the software business, is becoming less important, forcing the industry to confront a difficult question: if an AI agent can take a task from start to finish, then what exactly is the software company still selling?

That question sits at the center of the industry's next transition. Increasingly, the goal is no longer to help people move through a workflow more efficiently, but to absorb larger portions of the workflow itself. Companies across the industry are now placing significant bets on that possibility, and few illustrate the shift more clearly than Runway.

Runway's bet on execution

On May 13, Runway, the New York-based AI video company, launched Runway Agent, which it describes as an AI creative partner that can take a user from an idea to a finished, ready-to-publish video through a single conversation. A user can describe what they want, and the agent proposes a concept, develops story beats, creates the visual direction, and builds the final video across multiple scenes, complete with voiceover, dialogue, and music.

What makes this agent worthy of special mention is that until now, most AI video tools have only sped up individual parts of the production process by helping users generate clips, add audio, or edit footage. Runway says its agent handles the entire workflow from start to finish, combining scripting, video generation, editing, and sound design into one conversational system.

If the Agent launch revealed where the company believes creative software is heading, the announcement that followed revealed where it believes that future may arrive fastest.

A day later, the company announced that it was opening a Tokyo office and investing an initial $40M to expand in Japan, including hiring a Head of Japan to lead the effort. Even before establishing a major local presence, Japan had already become Runway's third-largest market across enterprise and self-serve customers, its fastest-growing self-serve market in Asia, and a country where enterprise customers grew 300% over the past year.

The two announcements point to the same larger strategy. Runway is not simply selling a video tool, but an entire production workflow, and it is expanding into a market where that model may find some of its earliest and most advanced users. Japan's creative industries, including anime, advertising, broadcasting, and gaming, depend heavily on high-volume and highly structured production pipelines.

Cristóbal Valenzuela, co-founder and co-CEO,  described Japan as "one of the most sophisticated creative industries in the world," arguing that the company's rapid growth there reflects that demand. The company's $40M investment is effectively a bet that creative teams will increasingly reorganize production around AI agents rather than around the traditional workflows those agents are designed to replace. And that is the larger significance of Runway's recent moves. The company is not simply introducing a new creative tool. It is betting that the workflow itself is becoming the product.

The same story everywhere

More importantly, Runway is not an outlier. The same transition is beginning to appear wherever AI systems are moving from assistance toward execution. What begins as software that helps a person complete a task gradually evolves into software that attempts to handle the entire workflow on its own.

In software development, for example, GitHub Copilot originally became popular as an AI assistant that suggested lines of code while developers continued writing the rest themselves. But the industry, according to TechCrunch, has since moved toward fully autonomous agents: systems that take a ticket and return a pull request, with the developer reviewing the difference rather than writing the code.

Cognition's Devin is the most prominent example, described by industry observers as pushing beyond assistance toward agency entirely, working on multi-step tasks with minimal human intervention and driving toward completion autonomously.

A similar transition is happening in enterprise software. Traditional automation tools were typically designed to speed up a single part of a business process, such as procurement or approvals. Agentic AI systems, however, are increasingly being built to manage much larger sections of an operation, coordinating planning, decision-making, and execution across multiple systems to achieve a business outcome rather than simply complete a scripted task.

The broader pattern has become increasingly clear since generative AI first entered the mainstream. A new capability emerges, companies turn it into a product, and the strongest players eventually push to control the entire workflow because that is where long-term value lies. Individual tools can be copied relatively easily. A workflow that has become deeply embedded in a company's operations is much harder to replace.

Once a company begins selling a workflow rather than a tool, another question follows naturally: how should it charge for it? The answer matters because the economics of software were built around a very different assumption: that people, rather than software, were doing the work.

For decades, enterprise software was largely sold through seat-based subscriptions, in which businesses paid per employee who used the product. When AI tools arrived, many companies added pricing based on credits, tokens, or API usage, but the underlying model remained largely the same.

The pricing fault line

However, even the new pricing model may not hold for long. According to IDC forecasts cited by CIO, by 2028, most software vendors are expected to move away from pure seat-based pricing and instead charge based on usage, business outcomes, or organizational impact. Analysts quoted in the report suggest that companies may gradually replace fixed software seats with more flexible pricing tied to the actual work an AI system performs.

The logic behind this shift is that if one AI system can do the work previously handled by multiple employees, charging per seat no longer reflects the value being delivered. Runway has not announced outcome-based pricing yet, and its agent product is sold as a feature of existing Pro and enterprise plans. But the product is built for a world where that conversation is unavoidable. Runway Agent does not save time within a workflow; it delivers the workflow's output. Pricing it by the credit would be analogous to pricing a contractor by the hour while he builds your house, then ignoring the fact that he finished in a week instead of three months.

Yet controlling a workflow can be extraordinarily valuable, which is precisely why competition around agentic systems is becoming so intense. According to reports, the company is now valued at $5.3B and added $40M in annual recurring revenue during the second quarter of 2026. Those are significant numbers, but they remain small compared to competitors like OpenAI and Alphabet, whose scale, funding, and computing infrastructure far exceed that of most startups.

Google's DeepMind, in particular, has spent years developing AI agents, robotics systems, and simulated environments, while its Veo video model competes directly with Runway's core business. Large technology companies also possess something smaller AI firms struggle to match: massive computing infrastructure capable of supporting rapid improvements in AI models.

The workflow test

Runway's challenge now is not proving that AI can generate videos. The technology has largely cleared that hurdle. The harder task is convincing companies to reorganize real creative work around AI agents rather than around the workflows they have spent years building.

Yet the larger question extends well beyond Runway. For decades, software companies created value by helping people move through workflows more efficiently. The workflow belonged to the user, while the software accelerated individual steps. What companies like Runway are betting on is that this arrangement is beginning to reverse. As AI systems become capable of taking a task from request to finished output, the value shifts away from the individual tool and toward the system that controls the workflow itself. The question facing the software industry has then shifted away from how to make workers more productive towards what happens when software increasingly becomes the worker.

Roko Pro Tip

Monday Poll

Bite-sized Brains

• Waymo recalls 4,000 robotaxis: Waymo issued a recall for nearly 4,000 robotaxis to stop them from driving into highway construction zones, a reminder that autonomous driving still has hard edge cases to solve.

• Anime meets the keyboard: IQUNIX released a Ghost in the Shell edition mechanical keyboard, a stylish nod to how deeply anime aesthetics now shape gaming hardware.

• AI compresses project timelines: A Deutsche Bank exec says AI is cutting tech project times from years to months, a concrete sign of how much faster enterprise work is moving.

Meme Of The Day

The Toolkit

• Assembly AI: Speech-to-text API that handles transcription, speaker detection, and audio intelligence for production apps. 

• Chroma: Open-source vector database built for AI apps, fast to set up and easy to scale for RAG and embeddings. 

• Continue: Open-source AI code assistant that plugs into VS Code and JetBrains with full control over models and context.

Rate This Edition