The AI Competition Is No Longer a Matter of "Anthropic vs OpenAI"

Writing for TechCrunch, Russell Brandom argues that the competitive structure of AI models has entered a stage that goes beyond company-vs-company rivalry. As model capabilities have improved, AI behavior now produces real-world political consequences, and responding to it will require collective action rather than individual companies acting alone.

This debate confronts us with a fundamental question: is it acceptable to leave AI safety and governance to each company's self-regulation? The argument is that now that model performance has crossed a certain threshold, no single company's efforts can manage the risks without industry-wide coordination mechanisms.

That said, what kind of collective action would actually be realistic remains unclear at this point. Regulation, industry standards, or perhaps an entirely new international framework — the discussion has only just begun.

Startup Lindy Fully Migrates From Claude to DeepSeek, Calling It "a Decision That Staked the Business's Survival"

AI startup Lindy has completely migrated from Claude (Anthropic) to DeepSeek. According to The Decoder, the direct trigger was AI costs exceeding personnel expenses, and CEO Flo Crivello described the move as "a judgment call made with the business's survival at stake." The migration reportedly delivered cost savings on the order of millions of dollars.

This case shows that AI cost pressure has entered a stage where it translates directly into real business decisions rather than remaining an abstract concern. That said, Lindy's case does not generalize to every company. Whether switching to DeepSeek succeeds or fails depends on multiple factors, including the nature of the tasks, latency requirements, and data governance policies. It should be understood as a single example of a cost-first decision that, "at this point in time," paid off.

Why Europe Wants "Its Own AI"

Steven Levy of WIRED reports on the growing momentum toward establishing AI sovereignty in Europe. The prevailing view remains that building a homegrown European top-tier model is still difficult, but the article characterizes the existence of the "Trump administration" as a tailwind for Europe. The dynamic at work: concern over American technological dominance is acting as a force that fosters unity within Europe.

The European Union (EU) has taken the lead on the regulatory front with instruments such as the AI Act, but it has lagged behind the US and China in model development. Achieving homegrown AI means overcoming barriers of investment, talent, and compute resources. Still, it is worth noting that the shifting geopolitical environment is generating unprecedented momentum.

The Day AI Coding Agents Cost More Than Developers

The Register reports that the cost of AI coding agents may soon surpass what human developers earn. AI-powered development has been touted as a trump card for productivity gains, but as autonomous, agent-style execution becomes the norm, the cumulative cost of API calls is ballooning beyond expectations.

Japan's developer community shares a similar sense of the problem. On Qiita, a hands-on report titled "After Breaking My AI Coding Agent 17 Times, I Finally Found a Solution" was published, sparking active discussion about how to design the balance between agent autonomy and control. Another article, "If You're Running AI Agents in Production, Data Product Design Decides Everything," argues that data quality and operational design make or break an agent's success. Beyond the technical excitement, it seems the operational groundwork is steadily being laid.

Simon Willison Reflects on "AI and Responsibility"

Simon Willison has published a reflection on where responsibility lies in AI systems. The question of who bears responsibility for harm caused by AI outputs — the developer, the provider, or the user — is growing in importance as AI becomes embedded in more and more decision-making.

Willison stresses the importance of distinguishing between the responsibility of humans who use AI tools as "instruments" and the problems lurking in the design of the tools themselves. This discussion connects to the need for "collective action" described above and to Europe's regulatory approach, offering a valuable perspective for thinking about the future direction of AI governance.

---