AI + Quantum

Highlights

Top Insights

AI may absorb some of quantum computing’s expected use cases, but not eliminate quantum’s role. AI systems are already tackling chemistry, materials science, biology, and weather forecasting through pattern recognition. That creates competitive pressure because AI can produce useful approximations without waiting for mature quantum hardware. But AI models still depend on training data and inferred patterns; quantum computers could eventually provide higher-fidelity simulations of physical reality.

AI and quantum computing are more likely to develop as complementary technologies than as substitutes. The strongest opportunities come from combining them in a feedback loop: quantum systems generate more accurate simulations, while AI helps identify where those costly simulations will create the most value.

Source: AI and quantum computers will be frenemies (The Economist)

Top News

1. Moonshot AI released Kimi K3 open weights.
2. Gemini Robotics 2 introduces adaptable AI models that give robots whole-body control and advanced dexterity.
3. xAI launched a Grok add-on for Google Workspace that works inside Sheets, Slides, and Docs.
4. Make added subagents and tool-output filtering to its AI Agent app for more modular automations.
5. Figma Make added properties-panel editing and annotations to make AI-assisted design-to-code work more controllable.

Additional Insights

1. From July Onward, Large AI Models Enter an ‘Infinite War’ (36Kr)
The article argues that China’s large-model industry has entered an “infinite war” with no settled business model or durable competitive moat. DeepSeek is betting on making intelligence cheap through open models and aggressive pricing; Kimi is betting that users will pay more for agents that complete entire tasks; ByteDance is turning AI into an emotional companion; Tencent is embedding it into social and commercial scenarios; and Alibaba is building a full-stack AI infrastructure spanning models, cloud services, and consumer applications. The deeper contest is therefore not simply over benchmark performance, but over how intelligence should be priced—by tokens, completed tasks, user attention, transactions, or cloud consumption. Because technical advantages can be copied quickly and switching costs remain low, the article concludes that no Chinese AI company yet has a true moat; however, that uncertainty also means the market remains open, and the eventual winner will shape how ordinary users access and pay for AI.

2. Research: How AI Agents Broaden the Scope of Knowledge Work (Harvard Business Review)
A field study comparing Perplexity’s conversational Search tool with its autonomous Computer agent suggests that agents do more than accelerate existing workflows—they expand what knowledge workers attempt. On matched tasks, the agent performed about 26 minutes of autonomous work versus 33 seconds for Search, reduced estimated completion time from 269 to 36 minutes and costs by 94%, and produced 55% fewer dissatisfied user responses. Because it could independently break down and execute multistep assignments, users shifted their attention toward reviewing, verifying, and extending outputs rather than manually coordinating each step. They also delegated more complex, cross-functional tasks requiring broader expertise and higher-order thinking, including activities rarely attempted with a conventional assistant. The central implication is that organizations should view agents not merely as productivity tools, but as a new form of delegation that can reshape roles, workflows, and the practical boundaries of knowledge work.

3. How is AI changing the skills for leadership and how should organizations prepare? (World Economic Forum)
AI is dismantling the traditional career ladder by automating many routine entry-level tasks that once helped employees build judgment, credibility and leadership experience. Although HR leaders increasingly expect junior roles to evolve into AI-supervision positions and see middle managers as essential to effective AI adoption, many organizations still lack adequate AI-focused training. The article argues that leadership must shift from technical expert and authority figure to human-centred enabler and orchestrator—someone who can identify new problems, provide context, exercise judgment, earn trust and align people and machines around shared goals. Organizations should therefore redesign work and career development, not merely job descriptions, creating deliberate learning loops in which employees guide, challenge and take accountability for AI-assisted work so they can develop the judgment and contextual skills future leaders will need.

4. Capable language models can outgrow the benefits of collaboration (Nature Machine Intelligence)
This study tests whether teams of AI agents outperform a strong single agent across 260 configurations, six benchmarks, five coordination structures, and three language-model families while holding prompts, tools, and compute budgets constant. Its central finding is that multi-agent collaboration is not inherently better: it tends to help when tasks can be cleanly divided and performed in parallel, but often hurts sequential or tightly interdependent tasks because coordination adds overhead, conflicting information, and amplified errors. The strongest predictor is the single agent’s baseline capability; once its performance exceeds an empirical threshold of roughly 45%, adding agents usually provides little benefit and may reduce accuracy—a rule that predicted collaboration’s effect in 94% of validation cases. The researchers’ broader predictive model selected the best architecture in 87% of held-out configurations, suggesting that agent-system design should be based on task structure and demonstrated baseline performance rather than the assumption that more agents automatically produce better results.