How Nooks scaled their AI platform 100x without breaking

The customer

Nooks is the “Agent Workspace” for sales, unifying a fragmented sales stack into one intelligent system. The company’s AI interface automates repetitive tasks like account research and cold outreach, so sales teams can focus on higher-leverage goals like strategy and building customer trust. This lets a team of any size rapidly multiply the output of an entire department. Today, Nooks’s customer base includes leading SaaS and AI companies such as HubSpot, Notion, Replit, Decagon, and Rippling.

The challenge

Traditional tools like SQS or GCP Pub/Sub worked well for simple async task execution, but when our workflows consist of 10–20 steps, we needed built-in state management, retry policies, and visibility that they couldn't provide.

As Nooks’ product usage grew tenfold, their lean Platform Team needed an execution engine that could scale without adding significant operational burden. Their use case called for a product that could handle two fundamentally different classes of work simultaneously:

• Short, user-triggered workflows. Latency-sensitive, real-time workflows triggered when sales reps complete actions like placing a call. These needed to validate data, generate transcripts, fetch metadata, and write to multiple external systems within a few seconds.
• Long-running, asynchronous ETL workflows. Large-scale pipelines that sync tens of millions of CRM records from Salesforce and HubSpot, transform them, and upsert them into Nooks’s internal database against rate-limited APIs.

Queue solutions like SQS and Pub/Sub worked well for the async workflows, but lacked capabilities such as a built-in UI to inspect what was running, native retry policies, state management, and Durable Execution.

The solution

It took less than a day to spin up a POC, and a few weeks to move everything and get it running in production.

Nooks discovered Temporal through word of mouth from engineers who had used it at other companies. When scaling challenges arrived, the team was struck by how quickly they could build a proof of concept.

Several factors stood out during evaluation:

• A working POC in under a day
• Highly scalable for asynchronous workloads
• Native TypeScript SDK
• An opinionated framework made it easier to get started
• The Temporal Cloud Startup Program, which provided free credits and eliminated the need for infrastructure ownership

They adopted Temporal for both critical use cases. Temporal’s Workflows / Child Workflows / Activities architecture mapped cleanly onto their pipeline design and enabled reliable scaling.

• Call logging pipeline. When a rep makes a phone call and wants to save that data, a Temporal Workflow fires immediately. It validates the call data, generates a transcript, creates translations, and fetches metadata about the contact. Then it writes everything to Nooks’s internal system and fans out to any external integrations the customer has configured. Each step is a discrete, independently retryable Temporal Activity, so a failure in one step doesn’t require re-running the entire chain.

• CRM sync pipelines. When a customer connects Salesforce or HubSpot, Nooks needs to import all of their CRM data and keep it continuously synchronized. There are two sync modes: historical sync (a one-time bulk import) and incremental sync (a recurring sync of recently modified records). The datasets are large, with peak loads reaching 40 million records, so a flat structure wouldn’t scale. Each sync mode is its own orchestration Workflow, which then fans out into Child Workflows, which then fan out into chunks to improve processing time.

  • Incremental syncs use Temporal Schedules with a BUFFER_ALL overlap policy. Without it, a slow sync still running when the next one fires would either be skipped or spawn a duplicate.

Together, these two pipelines form a core feedback loop driving intelligence across the Nooks platform. The call logging pipeline creates the data; the CRM sync pipeline moves it back to where sales teams live.

The results

The call logging pipeline has generated minimal failure reports, while also syncing hundreds of thousands of CRM records across rate-limited APIs and integrations with less-than-99% uptime. The reliability compounds across every system built on top of it.

With Temporal, Nooks’ two-person Platform Team has succeeded in growing their call volume 100x without overhauling core architecture. Temporal removed the operational burden of scaling, optimizing logic, and building a UI.

Beyond Child Workflows, which enable fan-out patterns across their pipelines, Nooks called several Temporal capabilities as especially critical:

1. State management, which lets them see exactly where a failure occurred and what the previous step’s inputs and outputs looked like.

2. Automatic retries, which handle the reality that third-party CRM APIs regularly return errors mid-sync and enforce hourly and daily rate limits that are difficult to manage without a durable execution layer.

That reliability has had a direct impact on engineering time. Before Temporal, the call logging pipeline required a permanently assigned engineer to triage failures. Today, with a self-recovering architecture in place, that same pipeline runs with roughly one to two hours of triage per week, freeing the broader team to ship new capabilities to larger customers rather than engineering around infrastructure problems.

The takeaways

What’s most impressive is how Temporal scales without our team having to put in extra effort, and that adoption has spread across teams and other product use cases entirely on its own.

What started as a two-person platform team has since grown, but Nooks’s story proves that lean teams can build big infrastructure when deliberate engineering meets the right tools.

The Platform Team’s design philosophy from the start was to break every process into the smallest safe, independent unit of work, and build it to last. The call logging pipeline, built to solve one specific problem, became the architectural standard for the company’s entire async and ETL infrastructure.

As their usage continues to hyperscale, from one million calls towards 100 million, they’ll continue optimizing their pipelines to deliver a production-ready experience for GTM teams.

These pipelines have become a cornerstone for much of Nooks’s asynchronous data and ETL work. I like to refer to projects like these as “foundational building blocks” – systems that grow with the company, scale over time, and set standards for future engineers to build upon.