How VEED.IO powers four million daily video workflows with Temporal

The customer

VEED.IO is an AI-powered online video creation platform that makes creating videos easy and accessible to everyone. Millions of creators and businesses around the world use our products to tell stories, create content, grow their audience, and more. They are backed by Sequoia, with a team of over 170 people around the world — and growing fast.

VEED’s product has evolved from a streamlined editor into a broad AI‑powered studio, with around three million videos exported every month. Users create and repurpose content with automated subtitles, avatars, and emerging generative features. Highlights include the AI Video Generator, which helps turn scripts or prompts into finished videos, and the AI Playground / AI Models, a space where people can try the latest generative models for video and images in one place. Under the hood, these capabilities run through long chains of background jobs that transform, render, and deliver assets at scale.

A TypeScript-first engineering culture

VEED’s engineering philosophy emphasizes simplicity and developer productivity. The team standardized on TypeScript across frontend and backend and deploys on Google Cloud Platform with a lean setup so engineers can stay focused on user experience rather than infrastructure.

This consistency lets frontend‑leaning engineers contribute to backend Workflows without changing languages or mental models.

The challenge

Rapid growth and an expanding AI surface area exposed limits in VEED’s homegrown event‑based pub/sub system for background jobs. Teams had to hand‑roll retries, error handling, and monitoring for each integration, which created uneven reliability and slowed down delivery. The pain was most visible with third‑party AI APIs — some reliable, many not. Without consistent automatic retries and durable state, transient failures leaked to users and eroded trust. Integrating a new model often meant rebuilding plumbing instead of composing existing pieces, turning quick wins into longer projects precisely when market interest was peaking.

Why Temporal?

Before discovering Temporal, VEED’s platform team evaluated several alternatives to their homegrown solution. They investigated Apache Airflow but found it wasn’t the right fit for their use case. Node-RED was considered but ultimately rejected. The team even contemplated doubling down on their custom approach — essentially building their own Temporal.

Each alternative presented trade-offs that didn’t align with VEED’s needs:

• Switching to Python or Go-based solutions would break their TypeScript-only principle
• Building in-house would require significant engineering resources they preferred to invest in product features
• Existing solutions lacked the flexibility to handle their complex, AI-driven workflows

The solution

The breakthrough came when VEED discovered Temporal offered everything they needed: robust workflow orchestration, native TypeScript support, and the flexibility to start with Temporal Cloud while keeping the option to self-host later.

VEED’s platform team took a measured approach to adoption. They began with a single, modest Workflow to understand Temporal’s patterns and best practices. This proof-of-concept phase revealed some initial learning curves: terminology differences and architectural patterns that differed from their previous system.

There were a couple of naming quirks we found a little tricky. What we would call one thing, it was different in the Temporal docs.

But as the team gained familiarity, they discovered Temporal’s true power: the ability to compose complex Workflows from simple, reusable Activities. This compositional approach transformed how VEED built features.

If we build a feature for generative AI — text to image, text to video — we built that feature in one department. And now we can use that anywhere in VEED just by being able to use that Activity as part of another Workflow.

VEED restructured their entire backend architecture around Temporal Workflows and Activities. Every third-party API call (whether for avatars, subtitles, stock media, or AI models) became a Temporal Activity with built-in retry logic and monitoring.

This architectural shift paid immediate dividends. When Google released their Veo 3 video model, VEED shipped support the same day. A week later, when ByteDance released C-Dance, VEED again had it in production within hours.

The results

The impact on development velocity was perhaps the most striking outcome. Features that would have taken five days to build and properly implement with error handling could now go live the same day a new AI model was announced.

We could now capture the hype from Twitter, we could capture everybody talking about this model and say, we’ve got it too. We’re right there, whereas five days later it was old news.

This speed didn’t come at the cost of reliability. Temporal’s automatic retry mechanisms smoothed over the rough edges of flaky third-party APIs, dramatically reducing error rates and improving user experience. VEED went from struggling with unreliable third-party APIs to handling over 4 million Temporal Activities daily with confidence.

Beyond the core improvements in reliability and speed, Temporal filled gaps in VEED’s monitoring and cost management infrastructure. The built-in observability meant teams could instantly see how many times they’d called each API provider.

We can quickly say, oh, we called it 100 times yesterday. $6 a time. That’s $600. Great!

This visibility proved invaluable for managing costs and optimizing API usage.

The cultural impact was equally significant. Frontend engineers who previously struggled with backend concepts could now contribute to workflow development. The clear separation between workflow orchestration and activity implementation made the codebase more approachable.

From a business perspective, the reliability improvements were transformative. As Muhammad Talha, Content Marketing Manager at VEED, explains:

When you rely on third-party providers, it's always a gamble. So making sure that everything stays up and running, and you know the product doesn’t go out is super important, especially for a company like us who have a ton of customers all over the world.

The takeaways

VEED’s journey from a fragile custom system to a robust Temporal-powered platform offers valuable lessons for any team building on top of third-party APIs or managing complex workflows.

Start small, but plan for scale

VEED’s incremental approach — starting with simple Temporal Workflows before tackling complex orchestrations — allowed them to build expertise gradually. They invested in training and appointed Temporal champions in each team to spread knowledge effectively. “Find champions for Temporal in each team,” Tom advises. “Make sure each team had someone who was really knowledgeable about Temporal that could act as a guide.”

Embrace composition over monoliths

By breaking down features into reusable Temporal Activities, VEED created a library of building blocks that accelerated future development. This modular approach meant new features could be assembled rather than built from scratch.

Developer experience drives adoption

VEED’s commitment to keeping everything in TypeScript, combined with Temporal’s developer-friendly abstractions, meant their team could focus on building features rather than fighting infrastructure. The result was happier developers shipping better products faster.

We can move quickly, but we can do things the right way. Before we would have had to make a trade-off between getting a feature out really quickly and it maybe having some bugs, or being very scrappy, not following best practice, or taking weeks to build something in the right way. Now we don't really have to make that trade-off anymore.