OpenAI Engineer on GPT-5.5's Productivity Boost

In a recent discussion, Aaron Friel, a Member of Technical Staff at OpenAI, shared his insights into the impact of GPT-5.5 on developer productivity. Friel, whose role at OpenAI involves engineering acceleration, highlighted how the new model is not only faster but also a powerful tool for modernizing and improving existing codebases. His experience offers a glimpse into how advanced AI is directly augmenting the work of software engineers.

Aaron Friel's Role at OpenAI

Aaron Friel is a key figure at OpenAI, focusing on engineering acceleration. His work involves ensuring that the organization's complex systems run smoothly and efficiently. This includes managing continuous integration systems and helping to keep all of OpenAI's projects moving forward. His hands-on experience with cutting-edge AI models like GPT-5.5 provides valuable perspective on their practical applications and benefits.

The full discussion can be found on OpenAI Youtube's YouTube channel.

GPT-5.5: A Leap in Developer Productivity

Friel expressed his enthusiasm for GPT-5.5, describing the experience of using it as incredible. He noted that the model's speed is a significant factor in its utility. He stated, "I have not expected the tidal wave of pull requests and changes in... as a result of engineers having so much intelligence at their fingertips." This suggests a broad adoption and positive impact across OpenAI's engineering teams.

He elaborated on how GPT-5.5 has been applied to various tasks, including reviving older code projects. "We've had engineers run this model and work on a single task for more than 40 hours at a time," Friel shared. He emphasized the model's ability to understand and improve code, stating, "It's been incredible to see the use of the Codex harness... to be able to see that used for developer productivity... building on ChatGPT." This indicates that GPT-5.5 is not just a tool for new development but also for maintaining and enhancing legacy systems.

Revitalizing Older Codebases

A particularly compelling aspect of Friel's discussion was the model's effectiveness in updating and improving older code. He explained, "We've revived those [code projects] and been able to run them on those tasks and see them all the way to completion where something that hasn't been building for more than, you know, 5 years, 10 years." This highlights GPT-5.5's capability to bridge the gap between older, potentially outdated code and current development standards.

Friel believes that GPT-5.5 is the best teacher for engineers learning new technologies. He stated, "I have never seen... GPT-5.5, and what you're trying to see here with like old parts of work that have been can be re-tackled... I've spent a lot of time reviving some old code projects of mine, ten, 15-year-old code projects that I had back from college or earlier and revived those." He further added, "It's been incredible being able to run them on those tasks and see them all the way to completion where something that hasn't been building for more than, you know, 5 years, 10 years."

The Future of AI-Assisted Development

Friel sees GPT-5.5 as a transformative tool for developers, enabling them to tackle complex problems more efficiently. He noted the broad impact across the company, stating, "we're not just seeing increases in contributions from engineers and researchers, but from everyone across the company is being uplifted by this new model and the ability to interrogate our code to make their own changes and suggest improvements or ship new product features." This suggests a democratizing effect of the AI, empowering a wider range of employees.

When asked for advice for developers still using older models like GPT-4, Friel emphasized the advantages of GPT-5.5. "I think that Codex remains the best teacher I've ever seen for teaching me how to use something new, whether it's a new technology, or even using apps and scale integrations and Codex to navigate the wealth of information that we have available," he shared. This underscores the model's potential as a learning and development accelerator for the entire organization.