Anthropic Joins the Coding Agent Competition… Will AI Job Replacement Accelerate? [Song Young-chan's Silicon Valley Focus]

Anthropic, considered a 'rival to OpenAI,' has unveiled an AI agent (assistant) specialized in coding. The focus was on significantly enhancing the proactivity and reasoning ability of AI compared to the previous model, allowing it to code independently for a longer period. Concerns are rising that the competition in AI development among U.S. tech companies is rapidly shifting towards coding agents, which have higher demand and are more advantageous for monetization, potentially accelerating AI job replacement.

On the 22nd (local time), Anthropic unveiled a new AI agent specialized in coding, 'Claude Opus 4.' The new model has significantly improved its ability to use tools such as web searches compared to the previous model. According to Anthropic, Claude Opus 4 outperformed existing coding-specialized models with a 72.5% accuracy rate in the program development benchmark 'SWE-Bench.' Jared Kaplan, co-founder and Chief Science Officer (CSO) of Anthropic, said, "As tasks become more complex, the risk of the model veering in unexpected directions increases, so we focused on solving this issue," adding, "We enabled users to delegate many tasks at once."

Since the end of last year, Anthropic has halted investments in AI chatbot development and instead focused on improving the ability to perform complex tasks. The primary focus areas have been scientific research and coding. As a result, the new model demonstrated the ability to continuously perform coding tasks for 7 hours without specific user commands. According to Anthropic, the new model also managed to play a Pokémon game continuously for up to 24 hours, compared to the previous 45 minutes.

This is the fourth time in the past week that a major U.S. tech company has unveiled a new coding-specialized AI agent. Starting with OpenAI's 'Codex' on the 16th, Microsoft (GitHub Copilot) on the 19th, and Google (Jules) on the 20th have all unveiled coding-specialized agents. As corporate clients focus on work efficiency and automation using AI, coding-specialized agents are deemed to have high demand. Particularly, big tech companies are moving to monetize AI by creating high-priced subscription plans and selling licenses based on these models. In fact, on the 20th, Google unveiled an 'AI Ultra' plan priced at $249.99 per month (approximately 350,000 KRW) at its largest annual event, the developer conference (I/O).

Concerns are growing that coding agents will significantly reduce jobs. It is known that major U.S. tech companies like Google, Meta, MS, and OpenAI are generating 20-30% of new code through AI. According to global job information company Indeed, the proportion of entry-level job postings in the software (SW) development field has drastically decreased from 3.2% in 2022 to 1.2% this year. A local tech industry insider said, "Compared to past AI models focused on text generation, coding-specialized AI agents will have a much greater impact on jobs," adding, "AI development companies are targeting the demand from companies looking to replace manpower with AI, so the competition will become even fiercer."

Silicon Valley = Song Young-chan Correspondent 0full@hankyung.com