W01 - Reviewing AI Developments in 2024

Looking back at changes in the AI field in 2024 from my perspective, I focus only on several leading companies and limit the scope to information distribution and coding.

First, OpenAI. Major turnover in the core team exposed challenges in corporate governance; the upside of the bet on pretraining appears to be shrinking, and a sustainable business model has not yet been established. It has repeatedly missed opportunities in niche markets: search was overtaken by Perplexity, and coding was surpassed by Claude. Even so, ChatGPT’s monthly active users are approaching 600 million, far ahead of the second place. This shows that in the consumer market, brand recognition outweighs technical differences—most users can’t tell which model is better, so the most efficient way to choose an AI product is to choose a brand.

Claude showed a clear strategic approach in 2024. Its base models focus on coding, the management team is stable, and it is backed by Amazon. After Sonnet 3.5’s release, it gained an absolute advantage among developers, and more developers means more model API calls. With the introduction of the MCP protocol, its developer-centered growth strategy became even more apparent. This approach resembles Stripe: capture developer mindshare first, then leverage that to unlock the broader commercial ecosystem.

Cursor and Devin are both dark horses in the coding space, validating that context is key. I haven’t tried Devin—it's mainly B2B and reportedly feels more futuristic. Cursor’s superior DX over Copilot comes down to its handling and interaction of context. Of course, in some respects it still falls short of humans. Its understanding of temporal and spatial context is insufficient—for example, when the same line of code undergoes different changes, the time context is an important analytic factor; without it, inferring relationships between edits is difficult. It also struggles with understanding multiple background relationships and long-range reasoning—such as the composition of a monorepo, which package manager is used, or which proprietary infrastructure libraries are introduced. These contextual details are developer commonsense that Cursor cannot yet grasp well.

Software could be the largest source of growth in the next era. Practitioners can feel that software production models and costs are on the verge of a major shift. Mobile internet produced content; the next era will produce software. Increasingly, people without CS backgrounds will choose to develop software themselves to solve problems. Just as traditional media gave way to individual creators where anyone can publish, traditional developers will be followed by self-developers where anyone can write software.