Free updates are available to those who stay informed through the Artificial Intelligence myFT Digest, delivered directly to subscribers’ inboxes.
Artificial intelligence is on track to surpass human capabilities in writing code as leading companies, including OpenAI, Anthropic, and Google, compete to release systems that are transforming the software industry. OpenAI, based in San Francisco, introduced a new suite of models this week that independent assessments deem among the most proficient in computer programming. These new GPT-4.1, o3, and o4-mini models excel in addressing programming challenges, with the latter two employing ‘reasoning’ to thoroughly process complex queries.
Additionally, OpenAI unveiled a free system on Wednesday named Codex CLI. This AI "agent" is designed to assist users with coding tasks using its models. These advancements align with recent initiatives from competitors such as Anthropic, Google, Meta, and several start-ups, which see coding as a prominent early application for large language models.
The focus on programming as a new frontier for AI systems highlights one of the most concrete examples of how the technology might revolutionize industries, with numerous software developers already integrating new models into their workflows. Kevin Weil, OpenAI’s chief product officer, remarked on the Overpowered podcast that this year marks the point when AI becomes superior to humans in competitive coding permanently. He compared this leap to AI’s triumph over humans in chess years ago, suggesting it holds a more democratizing potential by enabling widespread software creation.
Significant industry experts indicate that large language models (LLMs) have accelerated the software development process by generating entire code segments from basic textual instructions. These AI systems can also detect and attempt to rectify errors. Over the past year, AI models have significantly enhanced their ability to comprehend complex patterns, reason through programming issues, and solve them logically. For instance, in 2023, AI systems could solve only 4.4% of coding challenges on an industry test called SWE-bench; this figure surged to 69.1% this year. Concurrently, Microsoft’s platform GitHub revealed that 92% of US-based developers utilize AI coding tools.
Misha Laskin, co-founder and CEO of coding start-up Reflection AI, stated that AI coding is cutting significant costs for engineers, sometimes delivering services worth $10,000 on demand. The industry is witnessing an unprecedentedly large market expansion. Start-ups like Reflection have attracted considerable investor interest, having raised $130 million with funding from Sequoia and Lightspeed. Anysphere, a company behind the coding automation tool Cursor, secured $105 million at a $2.5 billion valuation in January.
Eiso Kant, co-founder of Poolside, a start-up that raised $500 million in October at a $3 billion valuation from investors including Bain Capital and Nvidia, said, "We are driving the cost down of what it means to do intelligent work, which means we’re going to have to rethink some of our roles."
Meta introduced a model called Code Llama last year, which utilizes text prompts to generate and discuss code. Anthropic launched its own coding product, Claude Code, in February. Mike Krieger, chief product officer at Anthropic, highlighted that the software engineer’s role will increasingly involve understanding user requirements, teamwork, and verifying that the final product meets expectations. He described the evolving role as a facilitator or conductor for these AI agents.
Thomas Wolf, co-founder of Hugging Face, an open-source AI platform, emphasized that coding will not disappear. Instead, coders will harness these tools to increase their efficiency.
Additional reporting was provided by George Hammond in San Francisco.