Vitaly Gordon started Salesforce Einstein in a basement with 5 people in 2016. It did not take too long for it to grow into an unequivocal success for Salesforce: improving internal company operations, used by over 10K customers, producing over 10 billion predictions every day, as well as cutting edge research, with hundreds of people working on it.
So why is Gordon not enjoying the fruits of his labor at Salesforce?
Because, as he put it, they were not practicing what they preach. Gordon realized that engineering teams in organizations are not at all data-driven as they should be. He left his role as VP, Data Science and Engineering at Salesforce Einstein and embarked on a quest to make software engineering data-driven, along with some of his former colleagues.
Faros AI is the company Gordon co-founded in 2019 to provide engineering teams with deep visibility into their operations so they can ship products faster. The Faros Engineering Operations Platform is already in use by the likes of Box, Coursera, and GoFundMe.
Faros AI today announced it has raised $16 million in seed funding led by SignalFire, Salesforce Ventures and Global Founders Capital with participation from seasoned tech luminaries including Maynard Webb, Frederic Kerrest, Adam Gross, and more.
What’s more, the company is also announcing the general availability of its free open-source Community Edition, Faros CE. We caught up with Gordon to discuss his journey with Faros AI, the philosophy of what they call EngOps, and the making of the Faros AI platform.
Analytics as the lighthouse of software engineering teams
Faros is Greek for lighthouse. As Gordon noted, marine-inspired analogies are going strong in the infrastructure space. It started with Docker, and then along came Kubernetes, which is Greek for a sea captain. So if Kubernetes is the helmsman that steers the ship, what points the way? That would be the lighthouse, and Faros AI wants to be the lighthouse.
Gordon refers to what Faros does as EngOps. If you’re familiar with DevOps, you may think that EngOps is similar — but it’s not. In reality, what Faros AI does can be summarized as analytics for software engineering teams. The reason Faros is using the term EngOps, Gordon said, is a nod to other disciplines.
Looking at roles such as Sales Operations, Marketing Operations, or Recruiting Operations, we find them filled in by highly analytical people. Their job is to get data from multiple sources, analyze the pipelines, find the bottlenecks, and then report to the relevant executives and work with them on improving what needs to be improved.
Faros AI is built around the notion of evangelizing that kind of role for software engineering. Gordon believes that every single company should have people who analyze data to advise engineering leads on allocating resources and making decisions.
You would think that with software engineering being entirely digital, with established practices and systems used, using analytics for this would have occurred to someone, and it would have been implemented already. Conceptually, it’s pretty straightforward, and Faros AI describes it using the Connect — Analyze — Customize triptych.
First, all the systems relevant for the software development process need to be connected, so their data can be ingested. Faros lets users connect systems such as code repositories, CI/CD, ticket management and project management software into one centralized system of record.
That is a prerequisite to being able to do analytics. It’s also not as simple as it sounds. Beyond getting the connectors in place, the data has to be integrated and aligned, and Gordon said it takes “some kind of intelligence” to stitch all those different data sources together. The goal is to trace changes from idea to production and beyond, incidents from discovery to recovery to resolution, and reconcile identities across the different systems.
Then comes the analysis, which is the core of the process. In Gordon’s experience, the metrics that are often used to measure developer productivity, such as lines of code or ticketing story points, may be easy to measure, but they are not really representative. If anything, Gordon said, there may be a reverse correlation between those metrics and the actual value generated.
In order to come up with what he claims can become a de facto set of metrics for software engineering, Gordon and his co-founders searched high and low. They came to rely heavily on DORA – Google Cloud’s DevOps Research and Assessment.
DORA studied over 1000 companies and measured over 100 metrics, using them to classify teams in 4 buckets — Elite, High, Medium and Low. They did that, Gordon said, based on metrics that focus on process and not people, measuring outcomes rather than outputs. This is the philosophy that Faros AI embraces as well.
Last but not least, customization allows Faros AI users to fine-tune metrics to their own needs and environment. As organizations differ in how they work and the environments they use, this is a necessary provision to ensure the platform works well for each scenario and the metrics collected reflect the reality on the ground.
Measuring and maximizing value
All that sounds fine and well, but how does it translate to tangible benefits in practice? To address this question, Gordon started by saying that just being able to see everything in one place is oftentimes enough to generate an “aha moment”. But it goes beyond that; he went on to add. One crucial aspect Faros AI has been able to help customers with is resource allocation:
“One of the things that we keep hearing from our customers, and it comes a lot from high-level management, or even sometimes the board, is: We hire more engineers, but we don’t seem to get more things done. Why is that? Especially in an environment where it’s so hard to hire more engineers, why don’t we see results?
One of the things we showed them is that if your bottleneck is not on engineers writing code, but in quality assurance, and you don’t have enough people there, then hiring more engineers to write more features will actually make things slower, not faster”, Gordon said.
Once organizations realized that, they responded by changing their hiring plans in order to address those bottlenecks, and that made a huge difference. Reassigning the existing workforce to address issues in the software engineering pipeline, rather than hiring more people, can result in the equivalent of hiring 20% more engineers as per Gordon.
The value comes not just from delivering software faster but also from improving software quality and minimizing downtime, Gordon went on to add. According to Google’s research, savings can be anywhere between $6 million and $250 million per year, depending on team size.
Faros AI is aimed at engineering team leads, CTOs and similar roles. While Gordon made a case for the value it can deliver to them; we wondered how the product is received by engineering team members, whose work is spotlighted. Experience with Faros AI customers shows that employee satisfaction goes up, Gordon said. That is because it reduces “internal bureaucracy”, resulting in a faster turnaround and having engineers see the impact of their work in the real world.
If talking about things such as software quality and value generated whets your appetite, you will have to manage your expectations. Trying to attribute the work of engineering teams to high-level business metrics is the holy grail for EngOps, Gordon said, but we’re not there yet.
The closest we can get at this point, he went on to add, is measuring how long it takes to get something to production. Given how engineering environments and systems sprawl, that’s not trivial. In Gordon’s experience, the Connect – Analyze – Customize cycle is something that many organizations do, under names such as developer productivity, engineering efficiency, or engineering empowerment.
Most of that work is completely undifferentiated, and it’s about infrastructure building. The thinking is that just like it makes sense for most organizations to use an off-the-shelf ERP or CRM system and customize it to their needs, EngOps should be no different.
For Gordon, Faros AI’s mission is to bring EngOps to as many organizations as possible. The release of Faros CE, the free, open-source Community Edition of the Faros AI platform, is an important step serving that goal. There are no real differences in capabilities between Faros CE and Faros AI Enterprise, except when it comes to features such as security and compliance, Gordon said.
Faros CE is a BI, API, and automation layer for all engineering operational data, including source control, task management, incident management and CI/CD data. It composes best-of-breed open-source software: Airbyte for data ingestion, Hasura for the API layer, Metabase for BI, and n8n for automation. Faros CE is container-based and is able to run in any environment, including the public cloud, with no external dependencies.
Faros AI Enterprise, available as SaaS with self-hosting options, will continue to be the monetization driver for Faros AI. However, Faros CE will also serve the goal of enabling customers to do things such as adding more connectors to their systems of choice. Faros AI worked in the reverse way companies sporting open source and enterprise versions typically do, starting with the enterprise version and then releasing the open source version.
This is also reflected in the way the company chose to fundraise, Gordon said. The seed round of $16 million comes after the company has been in operation for a while, with a fully functional platform and paying customers. This, Gordon went on to add, means that founders minimize the dilution of their stock and backers minimize their risk. The funding will be used to invest in the product, as well as grow the Faros AI team.