Major tech companies unveil AI agent guides

Technology giants have recently released comprehensive guides on building AI agents, marking a significant shift toward more autonomous systems that can execute complex workflows independently. These publications from Anthropic, OpenAI, McKinsey, and others provide crucial frameworks for developers looking to build next-generation AI systems that can act with minimal human intervention.

Five major tech companies have released AI agent guides in the past week, according to a LinkedIn post from former Streamlit and Snowflake employee Charly Wargnier . The post highlighted that "most are on point and practical," though one "raised eyebrows." These guides are particularly valuable for those "deep into agents and LLMs".

Anthropic released what was described as a "practical blueprint for building agentic workflows with Claude". This document gives developers concrete steps to leverage Claude's capabilities for creating systems that can work autonomously.

OpenAI published a 34-page guide focused on "building agents end to end" as well as additional guides on "AI agents in the enterprise" and "identifying and scaling use cases."

McKinsey contributed an 11-page perspective on "how AI agents will reshape enterprise ops." Though some critics found it "too top-down and light on real agent implementation," it provides valuable context on how McKinsey frames the conversation around AI agents.

Google also released a prompt engineering guide which, while not specifically agent-focused, offers insights that are "still super handy if you're building with LLMs".

Understanding AI agents

According to McKinsey's recent explainer published in March 2025, AI agents are "the tools we use to interact with AI" that can "automate and perform complex tasks, such as natural language processing, that would normally require humans."

Unlike conventional AI applications that simply provide information or content upon request, agents are systems that "independently accomplish tasks on your behalf." They possess unique characteristics that set them apart from simpler AI implementations:

1. They leverage an LLM to manage workflow execution and make decisions, recognizing when a workflow is complete and proactively correcting actions if needed.
2. They have access to various tools to interact with external systems—both to gather context and to take actions—and dynamically select appropriate tools depending on the workflow's current state.

Why AI agents matter now

The significance of these guides extends beyond technical documentation. They represent a fundamental shift in how AI systems are being developed and deployed. According to McKinsey, AI agents are "moving from thought to action" with major investments from Google, Microsoft, OpenAI and others in the past 18 months into "software libraries and frameworks to support agentic functionality."

Applications like Microsoft Copilot, Amazon Q, and Google's upcoming Project Astra, powered by large language models, are "making a shift from knowledge-based tools to ones that are more action based." McKinsey suggests that "in the near future, agents could become as commonplace as mobile applications."

The economic impact could be substantial. McKinsey research has estimated that generative AI's impact on productivity "could add trillions of dollars in value to the global economy", with AI agents playing a critical role in actualizing this potential.

Technical foundations of AI agents

OpenAI's practical guide breaks down agent design into three fundamental components:

1. Model: The LLM powering the agent's reasoning and decision-making
2. Tools: External functions or APIs the agent can use to take action
3. Instructions: Explicit guidelines and guardrails defining how the agent behaves

The guide emphasizes that different use cases require varying approaches to orchestration, from simple single-agent systems to more complex multi-agent architectures.

Enterprise applications and adoption

These guides are being released amid growing enterprise adoption of AI agents. Morgan Stanley, for example, has deployed an AI assistant using GPT-4 aimed at helping "thousands of financial advisors" quickly find and synthesize answers from their internal knowledge base.

Anthropic's "computer use" AI agent represents a significant advancement, as it "can transform office work by watching and learning from activity across different software programs." Unlike previous AI systems limited to specific applications, Anthropic's approach "lets its AI system operate computers just as humans do – by interpreting what's on screen and using the mouse and keyboard."

Such capabilities are driving competition between major tech companies. Anthropic's move "puts it in direct competition with tech giants Microsoft, Google, and OpenAI, who are all vying to automate routine computer tasks for businesses looking to boost productivity and cut costs."

Technical standards emerging

A significant development in the AI agent ecosystem is the emergence of standards for connecting these systems to external data sources and tools.

Anthropic's Model Context Protocol (MCP) has emerged as an important standard. MCP "simplifies artificial intelligence (AI) integrations by providing a secure, consistent way to connect AI agents with external tools and data sources."

In a notable industry development, OpenAI CEO Sam Altman announced in March 2025 that OpenAI will add support for Anthropic's MCP across its products, including the desktop app for ChatGPT. This represents an unusual case of cooperation between competing AI companies.

MCP "lets models draw data from sources like business tools and software to complete tasks, as well as from content repositories and app development environments." This functionality is crucial for AI agents to be effective in real-world business environments where data often resides in multiple systems.

Practical implementation tools

To facilitate development, both major AI companies have released specialized tools:

OpenAI's Agents SDK is "a lightweight Python framework" released in March 2025 that focuses on "creating multi-agent workflows with tracing and guardrails." The framework is "provider-agnostic and compatible with over 100 LLMs."

Anthropic launched the Model Context Protocol (MCP) in late November 2024, which is "designed to be model-agnostic, meaning any AI system—whether Claude, GPT-4, or open-source models—can implement it." Anthropic envisions MCP as "a USB-C port for AI, enabling seamless access to external knowledge and services."

For developers building agentic systems, these tools complement each other. "OpenAI's Agent SDK makes it easy to spin up agents, leveraging built in tooling, orchestration, and tracing from OpenAI. MCP makes it easy to access data from tools like databases, CRMs, etc."

The marketing industry impact

These developments have significant implications for the marketing industry, particularly in the realm of digital advertising. Digital marketers can leverage AI agents to automate and optimize complex advertising workflows.

In PPC (pay-per-click) advertising, tools like Anthropic's Model Context Protocol allow connecting real-time advertising data directly to AI agents. This enables capabilities such as identifying low-performing campaigns, recommending budget shifts, and offering optimization suggestions "with zero upfront data input."

For marketers using OpenAI's tools, GPT Actions provides similar functionality that allows AI to "request data (e.g., give me the last 30 days' campaign performance)" and "make changes (e.g., pause the campaign that has exceeded its target budget)."

These capabilities allow marketers to automate previously manual tasks, freeing up time for more strategic work. The technology is already making a significant impact, with GPT Actions being actively used to "link workflows with campaign performance data, analytics tools, and reporting dashboards."

Safety concerns emerging

As AI agents become more capable of taking autonomous actions, concerns about safety are also emerging. AI pioneer Geoffrey Hinton recently warned that AI agents present unique risks, stating "they now have these AI agents which are more dangerous than AI that just answers questions because they can do things in the world."

Among major AI companies, Hinton identified Anthropic as "the most concerned with safety," noting that many safety researchers who left OpenAI went to Anthropic. This underscores the importance of the guardrails and safety measures emphasized in the recently released agent guides.

Timeline of AI Agent developments

• March 2023: Anthropic releases Claude, an AI assistant designed to compete with ChatGPT
• June 2023: McKinsey research estimates generative AI could add trillions to global economy
• July 2024: McKinsey publishes "Why agents are the next frontier of generative AI"
• October 2022: Anthropic releases "computer use" AI agent, allowing Claude to operate computers by interpreting screens and using mouse/keyboard
• November 2024: Anthropic launches Model Context Protocol (MCP)
• January 2025: OpenAI releases Operator, allowing AI to automate tasks like planning vacations and filling out forms
• February 2025: OpenAI introduces Deep Research tool
• March 2025: OpenAI releases Agents SDK and announces support for Anthropic's MCP
• March 2025: McKinsey publishes "What is an AI agent" explainer
• March 2025: Amazon launches Nova Act AI model to compete with OpenAI and Anthropic
• April 2025: Major tech companies release comprehensive AI agent guides
• May 2025: Anthropic announces Integrations, connecting Claude to workplace applications