Oleg Klimov, Vlad Guber, and Oleg Kiyashko co-created Refact.ai in 2021 to encourage more organizations to adopt GenAI for coding by giving consumers greater flexibility and customization.
Klimov and Kiyashko built AI-based picture identification and security technologies for over a decade. Guber and Kiyashko were neighbors in Yuzhnoukrainsk, South Ukraine, from infancy.
“It was clear that AI would change the very notion of what engineering is,” Klimov emailed Eltrys. Being software engineers at heart, we determined we needed to be in the greatest position to survive it—creating an autonomous software engineering system.”
Many developers see AI-driven seismic developments in their field. Eighty-two percent of HackerRank poll respondents believe AI will “redefine” coding and software development.
In a 2023 poll by HeavyBit, 63% of programmers reported using GenAI. Employers are skeptical. In another study of company C-suite and IT workers, 85% worried about GenAI’s privacy and security.
Apple, Samsung, Goldman Sachs, Walmart, and Verizon have restricted internal GenAI use due to data security concerns.
Then how is Refact different? Klimov says on-premise.
Refact, like GitHub Copilot, Amazon CodeWhisperer, and other GenAI coding assistants, can answer natural language inquiries about code (e.g., “When was this dependency last updated?”), propose lines of code, and optimize its performance with a codebase.
Klimov suggested thinking of it as a “strong junior engineer” or a fake co-worker in a productive yet demanding team.
Refact does not require an internet connection, unlike most of its competitors. Klimov says it doesn’t upload basic telemetry.
Klimov added, “We’re developing better controls and processes around sources and uses of data, security, and privacy as we’re aware of the challenges [enterprises] face and want to ensure the integrity of our customers’ information and innovative breakthroughs.
Klimov said small, code-generating models trained on permissively licensed code power Refact’s platform, another competitive edge. Some IP experts believe that code-generating systems trained on copyrighted or otherwise restrictively licensed code may regurgitate that code when requested, putting organizations using them at risk.
GitHub and Amazon have implemented settings and regulations to reassure organizations worried about GenAI coding tool IP issues. They may not have made much progress. Nearly a third of Fortune 500 companies surveyed by Acrolinx in 2023 indicated intellectual property was their main concern regarding generative AI.
Klimov added, “We used permissive license code to train [our models] because our customers demanded it.”
Refact secured $2 million in investment from unknown investors and completed ~20 pilot projects with business customers due to its privacy- and IP-focused strategy. By this summer, Klimov expects the platform to make “a few million” a year in income from its cloud-hosted plan, which costs $10 per seat per month.
This is impressive because GitHub’s code-generating tools have struggled to earn a profit. Due to cloud processing costs, Copilot costs GitHub parent Microsoft up to $80 per user every month.
The London-based eight-person Refact team will soon upgrade Refact to run code autonomously, execute “multi-step” plans, and self-test code.
Klimov claimed they are developing a next-generation AI helper that can debug its own code and work on any huge codebase. We have sufficient internal funding to continue constructing the product. We’ve never had a lot of funding or venture capital, but what we have had is a lot of talented people who want to join the AI revolution and saw Refact as a place to thrive and develop something lasting.”