91制片厂

Manus 鈥� The world鈥檚 first autonomous agent

On March 6, 2025, Manus was launched. According to the parent company, Butterfly Effect, more than 2 million people have been waitlisted to receive the code.

On the Manus homepage, examples of agent-based applications are available for viewing; they demonstrate how Manus can assess the environment, select and execute tools, adapt, and deliver comprehensive results from a simple prompt.

ANALYST ANALYSIS

Manus is the latest innovation that has rivaled DeepSeek with respect to industry hype and interest from the AI industry. However, the innovation behind Manus is not due to a new LLM or any other newly developed offering disrupting the AI supply chain; instead, it鈥檚 a clever and smart use of existing AI technologies that demonstrates what is possible with AI agents 鈥� truly an autonomous agent.

Regarding comparisons being made between Manus and DeepSeek, it is unlikely that Manus will cause drastic movements in the stock market, nor give cause for organizations to reevaluate their investments in AI hardware as DeepSeek did. However, like DeepSeek, this does represent a transformation of what is possible with agent-based technologies and sets a benchmark that will have an impact on all organizations and vendors with their AI agent initiatives. The first vendor that needs to reassess their technologies will be OpenAI, since Manus appears to include the capabilities from both OpenAI Operator and Deep 91制片厂 into a single product, and it will likely be more price competitive when pricing becomes available.

OpenAI calls on the US government to ban DeepSeek

On March 13, 2025, OpenAI submitted a letter, a 15-page proposal, to the US Office of Science and Technology Policy outlining a regulatory framework that includes free rein to use copyrighted data to train its models and introduces increased export controls to ban the use equipment and models from China.

The proposal is the latest attack in OpenAI鈥檚 campaign against DeepSeek.

ANALYST ANALYSIS

The proposal is full of allegations without supporting facts, especially this statement: 鈥淒eepSeek is simultaneously state-subsidized, state-controlled, and freely available, the cost to its users is their privacy and security.鈥�

DeepSeek can be used in a private and secure manner, while OpenAI, the Internet version, is riddled with privacy and security exposures. DeepSeek is not a state-controlled model nor a distilled version of OpenAI (a previous allegation from OpenAI). This proposal is a desperate tactic to manipulate the government to ban a competitor under the guise of national security. If this is adopted, everyone loses except a handful of Big Tech vendors. OpenAI should focus on improving its competitiveness instead of crying foul when it starts to lose dominance due to innovation from its competitors.

UK Prime Minister to use AI to replace some work of civil servants

On March 13, 2025, the UK Prime Minister Keir Starmer delivered the quote, "No person's substantive time should be spent on a task where digital or AI can do it better, quicker, and to the same high quality and standard."

Starmer delivered this statement as part of the government鈥檚 new initiative focused on improving government efficiency through digital transformation and artificial intelligence.

ANALYST ANALYSIS

The UK is a world leader with AI, and earlier this year it announced its 10-year AI Opportunities Action Plan that included 50 recommendations to maintain its leadership position, grow the economy, and improve lives.

This recent announcement of transforming the public sector with AI has sent shockwaves through the public sector, with unions already warning Starmer that he must protect jobs instead of scapegoating civil servants. Starmer responded that he was not questioning the commitment of civil servants, but instead saying the system needs reform.

While an acknowledged leader in driving AI, the government has adopted a principle-based approach versus introducing legislation to address the potential risks with AI. It is currently attempting to reengineer its copyright laws to be more open and favorable for Big Tech vendors to train their models with copyrighted data.

Canada launches first-ever Artificial Intelligence Strategy for the federal public service

On March 4, 2025, the Canadian federal government announced its first-ever AI strategy for federal public sector. It is a comprehensive strategy that identifies several action items. Foundational deliverables include enabling a common infrastructure and establishing AI governance and AI risk management frameworks. Beyond foundational actions, the government is adopting a 鈥淭hink AI鈥� culture, where departments are challenged to identify key business problems that could be transformed using AI.

ANALYST ANALYSIS

In 2017, Canada became the world鈥檚 first country to announce a national AI strategy, which established Canda as an AI leader focused on AI research, talent development, and improving collaboration between academia, industry, and government.

This announcement leverages the AI ecosystem built by the national AI strategy to improve adoption of AI within the public sector in a responsible manner.

The strategy will advance four key priority areas: establishing an AI center of expertise to support and to help coordinate government-wide AI efforts; ensuring AI systems are secure and used responsibly; providing training and talent development pathways; and building trust through openness and transparency in how AI is used.

Anthropic announces Claude 3.7 Sonnet and Claude Code

On February 24, 2025, Anthropic announced the availability of Claude 3.7 Sonnet and Claude Code. Claude 3.7 Sonnet is Anthropic鈥� s most capable model and is positioned as a hybrid reasoning model. Claude Code is currently available as a limited research preview and is targeted for developers, enabling them to delegate tasks to Claude directly from their terminal. The product announcement continues to position Anthropic as one of the most advanced LLM providers, competing against OpenAI, Google Gemini, and DeepSeek.

ANALYST ANALYSIS

The Claude 3.7 Sonnet hybrid reasoning model refers to the model鈥檚 ability to adapt which reasoning strategy to use for a given prompt rather than have separate models for each reasoning strategy. In the hybrid model, one option is to respond quickly, similar to what most LLMs (like GPT-4) are using today. The second option, Thinking Mode, involves more 鈥渢hink time,鈥� an approach often referred to as 鈥淐hain-of-Thought鈥� (used by OpenAI o1/o3, Gemini and DeepSeek) reasoning for complex problems. In this mode, Claude 3.7 will allow developers to see the model鈥檚 step-by-step reasoning process.

Claude Code represents Anthropic鈥� s first agentic coding tool, built on top of Claude 3.7 Sonnet. As an agentic coding tool, it brings new capabilities for developers to leverage, including editing files, running tests, committing code, and interacting with command-line tools.