The founders who previously sold their livestreaming video startup Periscope to Twitter are back with a new startup — and no surprise, it’s an AI-focused company this time around.

On Wednesday, former Twitter head of product Kayvon Beykpour announced the launch of Macroscope, an AI system aimed at developers and product leaders that summarizes updates to a codebase and catches bugs, among other things.

The startup was co-founded by Beykpour, now Macroscope CEO, in July 2023, along with childhood friend Joe Bernstein, also previously of Periscope and their prior enterprise startup, Terriblyclever, which was sold to Blackboard in 2009. They’re joined by co-founder Rob Bishop, who sold his computer vision and machine learning company, Magic Pony Technology, to Twitter in 2016.

The company describes its product as an “AI-powered understanding engine” that’s designed to save engineers time, and the type of product the founders “wish we’d had” when building their earlier companies.

Today, engineers use a variety of tools to keep track of work, like JIRA, Linear, and spreadsheets, and spend too much time in meetings instead of building, Beykpour says. Macroscope is designed to fix this.

“I feel like I lived this pain…at every company I worked at, whether it was the startups that we built ourselves, or whether it was enormous public companies like Twitter, we sort of lived this problem the hard way,” Beykpour told TechCrunch in an interview.

“Trying to get a sense for what everyone was doing, especially when you have an organization like Twitter with thousands of engineers, it was literally most of my job — and my least favorite part of my job as the head of product at Twitter,” he said.

To address this issue and others, Macroscope’s customers first install its GitHub app, which gives the company access to the code base. They can then optionally install other integrations, like a Slack app, Linear app, and JIRA app. The software then does the rest of the work by analyzing the code and noting what’s changing.

This involves a process called code walking, which uses the Abstract Syntax Tree (AST) — a structural representation of programming code — to gather important context about how the customer’s code base works. That knowledge is then used in conjunction with large language models (LLMs).

Once up and running, engineers can use Macroscope to discover bugs to fix in their PRs (pull requests), summarize their PRs, get a summary of how the codebase is changing, and ask code research-based questions. Meanwhile, product leaders could use the software to get real-time summaries of product updates, productivity insights, answers to natural language questions about the product, code, or development activity, and more. This can help them determine what teams are prioritizing in terms of engineering allocation.

“You can ask natural language questions, regardless of what your technical ability is,” notes Beykpour. “This might be very useful if you’re trying to learn about the code base without distracting a senior engineer on your team. Very valuable. If you’re a CEO and you want to understand literally, ‘what did we get done this week?’, your options are either ask Macroscope or go distract some teammates,” he adds. “One is a lot more expensive than the other.”

While there isn’t a product that offers a direct competitor to all that Macroscope offers, it does compete in the code review space — where developers examine and test code changes before they’re implemented — with tools like CodeRabbit, Cursor Bugbot, Graphite Diamond, Greptile, and others. However, the company said when it ran its own internal benchmark of over 100 real-world bugs, its product caught 5% more bugs than the next-best tool. It also generated 75% fewer comments. (It shared its benchmark publicly in a blog post.)

The software costs $30 per active developer per month, starting at five seats, and offers enterprise pricing and custom integrations for larger businesses. It requires the use of GitHub Cloud. Ahead of its launch, a number of startups and larger firms have been using the product, including XMTP, Things, United Masters, Bilt, Class.com, Seed.com, ParkHub, A24 Labs, and others.

The San Francisco-based startup has a team of 20 and is backed by $30 million in Series A funding, which was closed in July and led by Michael Mignano at Lightspeed. Other investors include Adverb, Thrive Capital, and Google Ventures. To date, Macroscope has raised $40 million total.