Macroscope: AI Code Understanding & Bug Fixing

A New AI Startup Emerges from Periscope Founders

The individuals behind the former livestreaming video platform, Periscope – which was acquired by Twitter – are introducing a new venture. Unsurprisingly, this latest undertaking centers around the application of artificial intelligence.

Introducing Macroscope: An AI System for Developers

On Wednesday, Kayvon Beykpour, previously the head of product at Twitter, revealed the launch of Macroscope. This AI system is specifically designed for developers and product leaders, offering features like codebase update summarization and bug detection.

The Team Behind Macroscope

Macroscope was co-founded by Beykpour, who now serves as CEO, in July 2023. He is joined by his childhood friend, Joe Bernstein, also a former member of the Periscope team and co-founder of Terriblyclever, a company sold to Blackboard in 2009. Completing the founding team is Rob Bishop, whose computer vision and machine learning firm, Magic Pony Technology, was acquired by Twitter in 2016.

An "AI-Powered Understanding Engine"

The company positions its product as an “AI-powered understanding engine” intended to streamline workflows for engineers. It represents the type of tool the founders themselves desired during the development of their previous companies.

Addressing Inefficiencies in Engineering Workflows

Engineers currently rely on various tools, such as JIRA, Linear, and spreadsheets, to manage their tasks. However, Beykpour observes that this often leads to excessive time spent in meetings, detracting from actual development work. Macroscope is engineered to resolve this issue.

Personal Experience Drives Innovation

“I personally experienced this challenge at every organization I’ve been a part of,” Beykpour shared with TechCrunch. “Whether it was the startups we created or larger public companies like Twitter, we confronted this problem directly.”

He further explained that a significant portion of his role as head of product at Twitter involved simply understanding what thousands of engineers were working on – a task he found to be his least favorite.

How Macroscope Works: Integration and Analysis

To implement Macroscope, customers initially install its GitHub application, granting the company access to their codebase. Optional integrations with platforms like Slack, Linear, and JIRA can also be added. The software then autonomously analyzes the code and tracks changes.

Leveraging Code Walking and Large Language Models

This process utilizes “code walking,” which employs the abstract syntax tree (AST) – a structural representation of programming code – to gather crucial context about the codebase. This information is then combined with the capabilities of large language models (LLMs).

Features for Engineers and Product Leaders

Engineers can utilize Macroscope to identify bugs in pull requests, summarize those requests, gain insights into codebase modifications, and pose code-related questions. Product leaders can access real-time updates, productivity metrics, and answers to natural language queries concerning the product, code, or development activities. This facilitates informed decisions regarding engineering resource allocation.

Natural Language Queries and Accessibility

“Users can ask questions in natural language, regardless of their technical expertise,” Beykpour emphasized. “This is particularly beneficial for those seeking to understand the codebase without burdening senior engineers. It’s incredibly valuable. If you’re a CEO wanting to know ‘What did we accomplish this week?’, your choices are either to ask Macroscope or to interrupt your team,” he added. “One option is significantly more cost-effective.”

Competitive Landscape and Performance

While no single product directly replicates the full scope of Macroscope’s offerings, it does compete within the code review space with tools such as CodeRabbit, Cursor Bugbot, Graphite Diamond, Greptile, and others. However, the company’s internal benchmarks, involving over 100 real-world bugs, demonstrated that Macroscope detected 5% more bugs than the next-best tool and generated 75% fewer comments. (These results were published in a blog post.)

Pricing and Early Adoption

Macroscope is priced at $30 per active developer per month, with a minimum of five seats. Enterprise pricing and custom integrations are available for larger organizations. It requires the use of GitHub Cloud. Prior to its official launch, several startups and established companies have been utilizing the product, including XMTP, Things, United Masters, Bilt, Class.com, Seed.com, ParkHub, A24 Labs, and others.

Funding and Team Size

The San Francisco-based startup currently employs a team of 20 individuals and has secured $30 million in Series A funding, led by Michael Mignano at Lightspeed in July. Additional investors include Adverb, Thrive Capital, and Google Ventures. To date, Macroscope has raised a total of $40 million.