Agentic AI – 💡Tech News & Insights

Stop Feeding Your AI Generic Data: How to Build Intelligence That Understands Your Company

The future of enterprise AI: connecting intelligent systems to your proprietary knowledge.

In the executive suite, the conversation around Artificial Intelligence has shifted from “if” to “how.” We’ve all witnessed the power of generative AI, but many leaders are now asking the crucial follow-up question: “How do we make this work for our business, with our data, safely and effectively?” The answer lies in moving beyond generic AI and embracing a new paradigm that grounds AI in the reality of your enterprise. This is the world of Retrieval-Augmented Generation (RAG) and Agentic AI, and it’s not just the next step; it’s the quantum leap that transforms AI from a fascinating novelty into a strategic cornerstone of your business.

For C-level executives, the promise of AI is tantalizing: unprecedented efficiency, hyper-personalized customer experiences, and data-driven decisions made at the speed of thought. Yet, the reality has been fraught with challenges. Off-the-shelf AI models, while brilliant, are like a new hire with a stellar resume but no company knowledge. They lack context, can’t access your proprietary data, and sometimes, they confidently make things up, a phenomenon experts call “hallucination.” This is a non-starter for any serious business application.

This article will demystify the next generation of enterprise AI. We will explore how you can harness your most valuable asset, your decades of proprietary data, to create an AI that is not just intelligent, but wise in the ways of your business. We will cover:

The AI Reality Check: Why generic AI falls short in the enterprise.
RAG: Grounding AI in Your Business Reality: The technology that connects AI to your internal knowledge.
The Leap to Agentic AI: Moving from simple Q&A to AI that performs complex, multi-step tasks.
Real-World Implementation with Azure AI Search: A look at the technology making this possible today.
A C-Suite Playbook: Strategic considerations for implementing agentic AI in your organization.

The AI Reality Check: The Genius New Hire with No Onboarding

Imagine hiring the brightest mind from a top university. They can write, reason, and analyze with breathtaking speed. But on their first day, you ask them, “What were the key takeaways from our Q3 earnings call with investors?” or “Based on our internal research, which of our product lines has the highest customer satisfaction in the EMEA region?”

They would have no idea. They haven’t read your internal reports, they don’t have access to your sales data, and they certainly weren’t on your investor call. This is the exact position of a standard Large Language Model (LLM) like GPT-4 when deployed in an enterprise setting. These models are pre-trained on a massive, general, and publicly available dataset of text and code. They are masters of language and logic, but they are entirely ignorant of the unique, proprietary context of your business.

This leads to several critical business challenges:

Challenge	Business Impact
Lack of Context	AI-generated responses are generic and don’t reflect your company’s specific products, processes, or customer history.
Inability to Access Proprietary Data	The AI cannot answer questions about your internal sales figures, HR policies, or confidential research, limiting its usefulness for core business functions.
“Hallucinations” (Making Things Up)	When the AI doesn’t know the answer, it may generate a plausible-sounding but factually incorrect response, eroding trust and creating significant risk.
Outdated Information	The model’s knowledge is frozen at the time of its last training, so it is unaware of recent events, market shifts, or changes within your company.

Plugging a generic AI into your business invites inaccuracy and risk. The actual value is unlocked only when you can securely and reliably connect the reasoning power of these models to the rich, specific, and up-to-the-minute data that your organization has spent years creating.

RAG: Grounding AI in Your Business Reality

This is where Retrieval-Augmented Generation (RAG) comes in. In business terms, RAG is the onboarding process for your AI. It’s a framework that connects the AI model to your company’s knowledge bases before it generates a response. Instead of just relying on its pre-trained, general knowledge, the AI first “retrieves” relevant information from your trusted internal data sources.

Here’s how it works in a simplified, two-step process:

Retrieve: When a user asks a question (e.g., “What is our policy on parental leave?”), the system doesn’t immediately ask the AI to answer. Instead, it first searches your internal knowledge bases—like your HR SharePoint site, policy documents, and internal wikis—for the most relevant documents or passages related to “parental leave.”
Augment & Generate: The system then takes the user’s original question and “augments” it with the information it just retrieved. It presents both to the AI model with a prompt that essentially says, “Using the following information, answer this question.”

This simple but powerful shift fundamentally changes the game. The AI is no longer guessing; it’s reasoning based on your company’s own verified data. It’s the difference between asking a random person on the street for directions and asking a local who has the map open in front of them.

A diagram illustrating the RAG (Retrieval-Augmented Generation) architecture model, showing the flow between a client asking a question, semantic search, a vector database, and a large language model (LLM), with steps labeled from question to response.

A visual representation of the RAG architecture, showing how a user query is first enriched with data from a vector database before being sent to the LLM.

The Business Value and ROI of RAG

For executives, the implementation of RAG translates directly into tangible business value:

Drastically Improved Accuracy and Trust: By forcing the AI to base its answers on your internal documents, you minimize hallucinations and build user trust. Furthermore, modern RAG systems can provide citations, showing the user exactly which document the answer came from, creating an auditable trail of information.
Enhanced Employee Productivity: Imagine every employee having an expert assistant who has read every document in the company. Questions that once required digging through shared drives or asking colleagues are answered instantly and accurately. This frees up valuable time for more strategic work.
Hyper-Personalized Customer Service: When integrated with your CRM and support documentation, a RAG-powered chatbot can provide customers with answers tailored to their account history and the products they own, dramatically improving the customer experience.
Accelerated Onboarding and Training: New hires can get up to speed in record time by asking questions and receiving answers grounded in your company’s training materials, best practices, and internal processes.

The Next Evolution: From Smart Assistants to Proactive Digital Teammates with Agentic AI

If RAG gives your AI the ability to read and understand your company’s library, Agentic AI gives it the ability to act. An “agent” is an AI system that can understand a goal, break it down into a series of steps, execute those steps using various tools, and even self-correct along the way. It’s the difference between a Q&A chatbot and a true digital teammate.

Let’s go back to our earlier example:

A RAG-based query: “What were our Q3 sales in the EMEA region?” The system would retrieve the Q3 sales report and provide the answer.
An Agentic AI request: “Analyze our Q3 sales performance in EMEA compared to the US, identify the top 3 contributing factors for any discrepancies, draft an email to the regional heads summarizing the findings, and schedule a follow-up meeting.”

To fulfill this complex request, the agent would autonomously perform a series of actions:

Plan: Deconstruct the request into a multi-step plan.
Tool Use (Step 1): Access the sales database to retrieve Q3 sales data for both EMEA and the US.
Tool Use (Step 2): Analyze the data to identify discrepancies and potential contributing factors (e.g., marketing spend, new product launches, competitor activity).
Tool Use (Step 3): Draft a concise email summarizing the analysis, addressed to the appropriate regional heads.
Tool Use (Step 4): Access the corporate calendar system to find a suitable meeting time and send an invitation.

A flowchart illustrating the Agentic Retrieval-Augmented Generation (RAG) workflow, detailing the process from user query to response generation, including steps for memory, query decomposition, and search tool utilization.

An example of an agentic workflow, where the AI can plan, use tools, and even loop back to refine its approach if needed.

This is a paradigm shift. You are no longer just retrieving information; you are delegating outcomes. Agentic AI can orchestrate complex workflows, interact with different software systems (your CRM, ERP, databases, etc.), and work proactively to achieve a goal, much like a human employee.

Bringing it to Life: The Power of Azure AI Search

A screenshot of a chat interface for Azure OpenAI + AI Search, displaying a prompt to ask questions about data with example queries like 'What is included in my Northwind Health Plus plan that is not standard?'

The concepts of RAG and Agentic AI are not science fiction; they are being implemented today using powerful platforms like Azure AI Search. In the session at Microsoft Ignite, experts detailed how Azure AI Search is evolving to become the engine for these next-generation agentic knowledge bases. [1]

At the heart of this new approach is the concept of an Agentic Knowledge Base within Azure AI Search. This is a central control plane that orchestrates the entire process, from understanding the user’s intent to delivering a final, comprehensive answer or completing a task. Key capabilities highlighted include:

Query Planning: The system can take a complex or ambiguous user query and break it down into a series of logical search queries. For example, the question “Which of our products are best for a small business and what do they cost?” might be broken down into two separate queries: one to find products suitable for small businesses, and another to see their pricing.
Dynamic Source Selection: Not all information lives in one place. The agent can intelligently decide where to look for an answer. It might query your internal product database for pricing, search your SharePoint marketing site for product descriptions, and even search the public web for competitor comparisons—all as part of a single user request.
Iterative Retrieval: Sometimes, the first search doesn’t yield the best results. The new models within Azure AI Search can recognize when the initially retrieved information is insufficient to answer the user’s question. It can then automatically trigger a second, more refined search that takes into account what it learned from the first attempt. This iterative process mimics human research practices and yields more complete and accurate answers.

These capabilities, running on the secure and scalable Azure cloud, provide the foundation for building robust, enterprise-grade AI agents.

This is the example you can test and understand how it works: Azure OpenAI + AI Search

The Three Modes of Agentic Retrieval: Balancing Cost, Speed, and Intelligence

One of the most pragmatic aspects of Azure AI Search’s agentic knowledge base is the introduction of three distinct reasoning effort modes: minimal, low, and medium. This is a critical feature for executives because it allows you to dial in the right balance between cost, latency, and the depth of intelligence for different use cases.

Minimal Mode is the most straightforward and cost-effective option. In this mode, the system takes the user’s query and sends it directly to all configured knowledge sources without any query planning or decomposition. It’s a “broadcast” approach. This is ideal for scenarios where you are integrating the knowledge base as one tool among many in a larger agentic system, in which the agent itself already handles query planning. It’s also a good fit for simple, direct questions where the query is already well-formed and doesn’t require interpretation.

Low Mode introduces the power of query planning and dynamic source selection. The system will analyze the user’s query, break it down into multiple, more targeted search queries if needed, and then intelligently decide which knowledge sources are most likely to contain the answer. For example, if you ask, “What’s the best paint for bathroom walls and how does it compare to competitors?” the system might generate one query to search your internal product catalog and another to search the public web for competitor information. This mode strikes a balance between cost and capability, making it suitable for most production use cases that require intelligent retrieval without the overhead of iterative refinement.

Medium Mode is where the full power of agentic retrieval comes into play. In addition to query planning and source selection, medium mode introduces iterative retrieval. The system uses a specialized model, often referred to as a “semantic classifier,” to evaluate the quality and completeness of the retrieved results. It asks itself two critical questions: “Do I have enough information to answer the user’s question comprehensively?” and “Is there at least one high-quality, relevant document to anchor my response?” Suppose the answer to either question is no. In that case, the system will automatically initiate a second retrieval cycle, this time with refined queries based on what it learned from the first attempt. This mode is best suited for complex, multi-faceted questions where accuracy and completeness are paramount, even if it means a slightly higher cost and latency.

Understanding these modes is crucial for strategic deployment. You wouldn’t use a Formula 1 race car for a grocery run, and similarly, you don’t need the full power of medium mode for every query. By thoughtfully mapping your use cases to the appropriate retrieval mode, you can optimize both performance and cost.

A C-Suite Playbook for Adopting Agentic AI

For business leaders, the journey into agentic AI requires a strategic approach. This is not just an IT project; it is a fundamental transformation of how work gets done.

Start with Your Data Estate: The intelligence of your AI is directly proportional to the quality and accessibility of your data. Begin by identifying your key knowledge repositories. Where does your most valuable proprietary information live? Is it in structured databases, SharePoint sites, shared drives, or PDFs? A successful agentic AI strategy begins with a strong data governance and knowledge management foundation.
Focus on High-Value, High-Impact Use Cases: Don’t try to boil the ocean. Identify specific business problems where AI can deliver a clear and measurable return on investment. Good starting points often involve:
- Internal Knowledge & Expertise: Automating responses to common questions from employees in HR, IT, or finance.
- Complex Customer Support: Handling multi-step customer inquiries that require information from different systems.
- Data Analysis and Reporting: Automating the generation of routine reports and summaries from business data.
Embrace a “Human-in-the-Loop” Philosophy: In the early stages, it’s crucial to have human oversight. Implement systems that allow a human to review and approve the AI’s actions, especially for critical tasks. This builds trust, ensures quality, and provides a valuable feedback loop for improving the AI’s performance over time.
Partner with the Right Experts: Building agentic AI systems requires a blend of skills in data science, software engineering, and business process analysis. Partner with teams, either internal or external, who have demonstrated expertise in building these complex systems on enterprise-grade platforms.
Measure, Iterate, and Scale: Define clear metrics for success. Are you reducing the time it takes to answer customer inquiries? Are you increasing employee satisfaction? Are you automating a certain number of manual tasks? Continuously measure your progress against these metrics, use the insights to refine your approach, and then scale your successes across the organization.
Prioritize Security and Compliance from Day One: When your AI is accessing your most sensitive business data, security cannot be an afterthought. Ensure that your agentic AI platform adheres to your organization’s security policies and industry regulations. Key considerations include:
- Data Encryption: Both data at rest and data in transit must be encrypted.
- Access Control: Implement robust role-based access control (RBAC) to ensure the AI accesses only the data the user is authorized to see. If a user doesn’t have permission to view a specific SharePoint folder, the AI shouldn’t be able to retrieve information from it on their behalf.
- Audit Trails: Maintain comprehensive logs of all AI interactions and data access for compliance and security auditing.
- Data Residency: Understand where your data is being processed and stored, mainly if you operate in regions with strict data sovereignty laws.

Financial Services: Intelligent Compliance and Risk Management

In the highly regulated world of finance, staying compliant with ever-changing regulations is a constant challenge. A significant investment bank implemented an agentic AI system that continuously monitors regulatory updates from multiple sources (government websites, industry publications, internal legal memos). When a new regulation is published, the agent automatically:

Retrieves the full text of the regulation.
Analyzes it to identify which business units and processes are affected.
Searches the bank’s internal policy database to find existing policies that may need to be updated.
Generates a draft impact assessment report for the compliance team.
Schedules a review meeting with the relevant stakeholders.

This system has reduced the time to identify and respond to new regulatory requirements by over 60%, significantly lowering compliance risk and freeing up the legal and compliance teams to focus on strategic advisory work.

Healthcare: Accelerating Clinical Decision Support

An extensive hospital network deployed a RAG-based clinical decision support system for its emergency department physicians. When a physician is treating a patient with a complex or rare condition, they can query the system with the patient’s symptoms, medical history, and test results. The system:

Searches the hospital’s internal database of anonymized patient records to find similar cases and their outcomes.
Retrieves relevant sections from the latest medical research papers and clinical guidelines.
Cross-references the patient’s current medications with known drug interactions.
Presents the physician with a synthesized summary, including treatment options that have been successful in similar cases, potential risks, and citations to the source data.

This has not only improved the speed and accuracy of diagnoses but has also served as a powerful continuing education tool, keeping physicians up-to-date with the latest medical knowledge without requiring them to spend hours reading journals.

Manufacturing: Predictive Maintenance and Supply Chain Optimization

A global manufacturing company integrated an agentic AI system into its operations management platform. The agent continuously monitors data from IoT sensors on the factory floor, supply chain logistics systems, and external market data. When it detects an anomaly—such as a machine showing early signs of wear or a potential disruption in the supply of a critical component—it autonomously:

Retrieves the maintenance history and specifications for the affected machine.
Searches the inventory system for replacement parts and identifies alternative suppliers if needed.
Analyzes the production schedule to determine the optimal time for maintenance with minimal disruption.
Generates a work order for the maintenance team and, if necessary, initiates a purchase order for parts.
Sends a notification to the operations manager with a summary and recommended actions.

This proactive approach has reduced unplanned downtime by 40% and optimized inventory levels, resulting in significant cost savings.

Retail: Hyper-Personalized Customer Experiences

A leading e-commerce retailer uses an agentic AI system to power its customer service chatbot. Unlike traditional chatbots that follow rigid scripts, this agent can:

Access the customer’s complete purchase history, browsing behavior, and past support interactions.
Retrieve product information, inventory levels, and shipping details from the company’s databases.
Search the knowledge base for troubleshooting guides and FAQs.
Suppose the customer has a complex issue (e.g., a defective product). In that case, the agent can autonomously initiate a return, issue a refund or replacement, and even suggest alternative products based on the customer’s preferences.

The result is a customer service experience that feels genuinely personalized and efficient, leading to a 25% increase in customer satisfaction scores and a significant reduction in the workload on human customer service representatives.

The “Black Box” Problem: Explainability and Trust

One of the most common concerns about AI is that it operates as a “black box”; you get an answer, but you don’t know how it arrived at that conclusion. This is particularly problematic in regulated industries or high-stakes decisions. The good news is that modern RAG systems are inherently more explainable than traditional AI. Because the system retrieves specific documents or data points before generating an answer, it can provide citations. You can see exactly which internal document or data source the AI used to formulate its response. This traceability is crucial for building trust and ensuring accountability.

However, it’s important to note that while you can see what data the AI used, understanding how it reasoned with that data to arrive at a specific conclusion can still be opaque, especially with the most advanced models. This is an active area of research, and as a business leader, you should demand transparency from your AI vendors and prioritize platforms that offer the highest degree of explainability for your use case.

Data Privacy and Ethical Use

When your AI has access to vast amounts of internal data, including potentially sensitive information about employees and customers, data privacy and ethical use become paramount. You must establish clear policies on:

What data the AI can access: Not all data should be available to all AI systems. Implement strict access controls.
How the AI can use that data: Define acceptable use cases and prohibit its use in ways that could be discriminatory or harmful.
Data retention and deletion: Ensure that data used by the AI is subject to the same retention and deletion policies as other company data.
Transparency with stakeholders: Be transparent with employees and customers about how AI is being used and what data it has access to.

Building an ethical AI framework is not just about compliance; it’s about building trust with your stakeholders and ensuring that your AI initiatives align with your company’s values.

The Strategic Imperative: Why Now is the Time to Act

The window of competitive advantage is narrowing. Early adopters of agentic AI are already seeing measurable gains in efficiency, customer satisfaction, and innovation. As these technologies become more accessible and the platforms more mature, the question is no longer “Should we invest in agentic AI?” but “How quickly can we deploy it effectively?”

Consider the following strategic imperatives:

First-Mover Advantage: In many industries, the companies that successfully integrate agentic AI first will set the standard for customer experience and operational efficiency, making it harder for competitors to catch up.
Data as a Moat: Your proprietary data is a unique asset that competitors cannot replicate. By building AI systems that are deeply integrated with your data, you create a sustainable competitive advantage.
Talent Attraction and Retention: Top talent, especially in technical fields, wants to work with cutting-edge technology. Demonstrating a commitment to AI innovation can be a powerful tool for attracting and retaining the best people.
Regulatory Preparedness: As AI becomes more prevalent, regulatory scrutiny will increase. Companies that have already established robust AI governance frameworks and ethical use policies will be better positioned to navigate the evolving regulatory landscape.

The Future is Now

The era of generic AI is over. The competitive advantage of the next decade will be defined by how effectively organizations can infuse the power of AI with their own unique, proprietary data and business processes. Retrieval-Augmented Generation (RAG) and Agentic AI are the keys to unlocking this potential.

By building AI systems grounded in your reality and capable of intelligent action, you are not just adopting a new technology; you are building a digital workforce that can augment and amplify your human team’s capabilities on an unprecedented scale.

Further Resources:

Microsoft Ignite Session: For a deeper technical dive, watch the whole session, “Build agents with knowledge, agentic RAG and Azure AI Search.”
GitHub Repository: Explore the session’s code and notebooks on GitHub.
Live Demo Example: See a simple RAG application in action, built on Azure AI Search: Azure Chat Demo.

Sources

[1] Fox, P., & Gotteiner, M. (2025). Build agents with knowledge, agentic RAG, and Azure AI Search. Microsoft Ignite. Retrieved from https://ignite.microsoft.com/en-US/sessions/BRK193?source=sessions

From Co-Pilot to Autopilot: The Evolution of Agentic AI Systems

Imagine a world where your digital assistant doesn’t just follow your commands, but anticipates your needs, plans complex tasks, and executes them with minimal human intervention. Picture an AI that can, when asked to ‘build a website,’ independently generate the code, design the layout, and launch a functional site in minutes. This isn’t a scene from a distant science fiction future; it’s the rapidly approaching reality of agentic AI systems. In early 2023, the world witnessed a glimpse of this potential when AutoGPT, an experimental autonomous AI agent, reportedly accomplished such a feat, constructing a basic website autonomously. This marked a significant leap from AI as a mere assistant to AI as an independent actor.

Agentic AI refers to artificial intelligence systems with agency—the capacity to make decisions and act autonomously to achieve specific goals. These systems are designed to perceive their environment, process information, make choices, and execute tasks, often learning and adapting as they go. They represent a paradigm shift from earlier AI models that primarily responded to direct human input.

This article will embark on a journey to trace the evolution of artificial intelligence, from its role as a helpful ‘co-pilot’ augmenting human capabilities to its emergence as an ‘autopilot’ system capable of navigating and executing complex operational cycles with decreasing reliance on human guidance. We will explore the pivotal milestones and technological breakthroughs that have paved the way for this transformation. We’ll delve into real-world applications and examine prominent examples of agentic AI, including innovative systems like Manus AI, which exemplify the cutting edge of this field. Furthermore, we will analyze the profound benefits these advancements offer, the inherent challenges and risks they pose, and the potential future trajectories of agentic AI development.

Our exploration will begin by examining the history of AI assistance, moving through digital co-pilot development, and then focusing on the key characteristics and technologies defining modern autonomous AI agents. We will then consider the societal implications and the ongoing dialogue surrounding the ethical and practical considerations of increasingly autonomous AI. Join us as we navigate the fascinating landscape of agentic AI and contemplate its transformative impact on our world.

Agentic AI: What Is It?

Agentic AI refers to artificial intelligence systems designed and developed to act and make decisions autonomously. These systems can perform complex, multi-step tasks in pursuit of defined goals, with limited to no human supervision and intervention.

Agentic AI combines the flexibility and generative capabilities of Large Language Models (LLMs) such as Claude, DeepSeek-R1, Gemini, etc., with the accuracy of conventional software programming.

Agentic AI acts autonomously by leveraging technologies such as Natural Language Processing (NLP), Reinforcement learning (RL), Machine Learning (ML) algorithms, and knowledge representation and reasoning (KR).

Compared to generative AI, which is more reactive to a user’s input, agentic AI is more proactive. These agents can adapt to changes in their environments because they have the “agency” to do so, i.e., make decisions based on their context analysis.

From Assistants to Agents: A Brief History of “Co-Pilots”

The journey towards sophisticated Artificial Intelligence agents, capable of autonomous decision-making and action, has its roots in simpler assistive technologies. The concept of an AI “assistant” designed to aid humans in various tasks has been a staple of technological aspiration for decades. Early iterations, while groundbreaking for their time, were often limited in scope and operated based on pre-programmed scripts or rules rather than genuine understanding or learning capabilities.

Think back to the animated paperclip, Clippy, a familiar sight for Microsoft Office users in the 1990s. Clippy would offer suggestions based on the user’s activity, which would be a rudimentary form of assistance. While perhaps endearing to some, Clippy’s intelligence was not adaptive; it lacked the capacity for learning or genuine autonomy. Similarly, early expert systems and chatbots could simulate conversation or provide advice within narrowly defined domains, but their functionality was constrained by the if-then rules hardcoded by their programmers. These early systems were tools, helpful in their specific contexts, but far from the dynamic, learning-capable AI we see today.

The Era of Digital Co-Pilots Begins

A significant leap occurred in the 2010s with the advent and popularization of smartphone voice assistants. Apple’s Siri, launched in 2011, followed by Google Assistant, Amazon’s Alexa, and Microsoft’s Cortana, brought natural language interaction with AI into the mainstream. Users could now verbally request information, set reminders, or control smart home devices. These assistants were powered by advancements in speech recognition and the nascent stages of natural language understanding. However, they remained largely reactive, responding to specific commands or questions within a predefined set of capabilities. They did not autonomously pursue goals or string together complex, unprompted actions.

In parallel, the software development sphere witnessed the emergence of AI code assistants, marking a more direct realization of the “co-pilot” concept in AI. A pivotal moment was the introduction of GitHub Copilot in 2021. Developed through a collaboration between OpenAI and GitHub (a Microsoft subsidiary), GitHub Copilot was aptly termed “Your AI pair programmer.” Leveraging an advanced AI model, OpenAI Codex (a descendant of the GPT-3 language model), it provided real-time code suggestions. It could generate entire functions within a developer’s integrated development environment (IDE). As a developer typed a comment or initiated a line of code, Copilot would offer completions or alternative solutions, akin to an exceptionally advanced autocomplete feature. This innovation dramatically enhanced productivity, allowing developers to generate boilerplate code and receive instant suggestions quickly. However, GitHub Copilot functioned as an assistant, not an autonomous entity. The human developer remained the pilot, guiding the process, while the AI served as the co-pilot, offering support and executing specific, directed tasks. The human reviewed, accepted, or rejected the AI’s suggestions, maintaining ultimate control.

The success of GitHub Copilot spurred a wave of “copilot” branding across the tech industry. Microsoft, for instance, extended this concept to its Microsoft 365 Copilot for Office applications, Power Platform Copilot, and even Windows Copilot. These tools, often powered by OpenAI’s GPT models, aimed to assist users in tasks like drafting emails, summarizing documents, and generating formulas. The term “co-pilot” effectively captured the essence of this human-AI interaction: the AI assists, but the human directs. These early co-pilot systems were not designed to initiate tasks independently or operate outside the bounds of human-defined objectives and prompts.

Co-Pilot vs. Autopilot – What’s the Difference in AI?

Understanding the distinction between a “co-pilot” AI and an “autopilot” AI is crucial to appreciating the trajectory of AI development. As we’ve seen, co-pilot AI systems, such as early voice assistants or coding assistants like GitHub Copilot, are designed to assist a human user in performing a task. They respond to prompts, offer suggestions, and execute commands under human supervision.

In stark contrast, an autonomous agent, the “autopilot” in our analogy, can take a high-level goal and independently devise and execute a series of steps to achieve it, requiring minimal, if any, further human input. As one Microsoft AI expert aptly put it, these agents are like layers built on top of foundational language models. They can observe, collect information, formulate a plan of action, and then, if permitted, execute that plan autonomously. The defining characteristic of agentic AI is its degree of self-direction. A user might provide a broad objective, and the agent autonomously navigates the complexities of achieving it. This is akin to an airplane’s autopilot system, where the human pilot sets the destination and altitude, and the system manages the intricate, moment-to-moment controls to maintain the course.

This significant leap from a reactive assistant to a proactive, goal-oriented agent has only become feasible in recent years. This progress is mainly attributable to substantial advancements in AI’s capacity to comprehend context, retain information across interactions (memory), and engage in reasoning processes that span multiple steps or stages.

Key Milestones on the Road to Autonomy

Critical AI research and technology breakthroughs have paved the path from rudimentary rule-based assistants to sophisticated autonomous agents. Let’s highlight some of the pivotal milestones and innovations that have enabled the development of increasingly agentic AI systems:

Rule-Based Agents and Expert Systems (1980s–1990s): These early AI programs, often called intelligent agents, operated based on predefined rules. They could perform limited, specific tasks like monitoring stock prices or filtering emails. While they laid the conceptual groundwork for software agents, their intelligence was derived from explicitly programmed logic, making them brittle and narrowly applicable. They set the stage conceptually for software “agents” but lacked accurate intelligence or autonomy.
Reinforcement Learning and Game Agents (2010s): A significant leap in agent capability emerged from reinforcement learning (RL). In RL, an AI agent learns through trial and error, optimizing its actions to maximize a cumulative reward within a given environment. DeepMind’s AlphaGo, which in 2016 demonstrated superhuman performance in the complex board game Go, and OpenAI Five, which achieved similar feats in the video game Dota 2 by 2018, showcased the power of RL. These systems were undeniably agents; they perceived their environment (the game state) and took actions (game moves) to achieve clearly defined goals (winning the game). However, their agency was highly specialized, meticulously tuned to a single task, and they could not interact using natural language or address arbitrary real-world objectives.
Transformer Models and Language Understanding (late 2010s): Google researchers’ introduction of the Transformer neural network architecture in 2017 marked a watershed moment for natural language AI. Models like BERT (Bidirectional Encoder Representations from Transformers) and GPT-2 (Generative Pre-trained Transformer 2) demonstrated astonishing improvements in understanding and generating human-like text. By 2020, OpenAI’s GPT-3, with its staggering 175 billion parameters, showcased an unprecedented ability to perform various language tasks—from writing essays and answering complex questions to generating code—often without task-specific training. This was a general-purpose language engine, and it hinted at the possibility that a sufficiently robust model could be adapted into an “agent” simply by instructing it in plain English.
The GitHub Copilot Launch (2021) signaled that assistive AI was emerging. As previously described, GitHub Copilot utilizes a fine-tuned GPT model (Codex) version to provide live coding assistance directly within a developer’s environment. It was one of the first instances where an AI was integrated as a “pair programmer” into a widely adopted professional tool. This demonstrated that large language models could serve as valuable teammates, not merely as clever chatbots, further solidifying the co-pilot paradigm.
Large Language Models Everywhere (2022): 2022 witnessed an explosion in LLMs’ application and public awareness. Based on OpenAI’s GPT-3.5 model, ChatGPT was released to the public in late 2022 and rapidly amassed over 100 million users. It provided an eerily capable conversational assistant for an almost limitless range of tasks that could be described in natural language. ChatGPT could draft emails, brainstorm ideas, explain intricate concepts, and, significantly, write functional code. Users quickly discovered that through conversational interaction, they could guide ChatGPT to achieve multi-step results, for example, “first brainstorm a story plot, then write the story, and now critique it.” However, the user still needed to guide each step explicitly. This widespread interaction led researchers and developers to ponder a crucial question: What if the AI could guide itself through these steps?
Tool Use and Plugins (2023): A critical enabling factor for the transition towards autonomous agents was granting LLMs the ability to use tools and perform actions beyond simple text generation. For example, OpenAI’s ChatGPT Plugins and Function Calling allowed the LLM to interact with external APIs, extending its capabilities beyond text manipulation. This meant the AI could, for instance, access real-time information from the internet, perform calculations, or even interact with other software systems. This development was pivotal in transforming LLMs from sophisticated text generators into more versatile agents capable of performing complex tasks.
AutoGPT and the Rise of Autonomous LLM Agents (2023): With tool-use capabilities established, enterprising developers rapidly pushed the boundaries of AI autonomy. In April 2023, an open-source project named AutoGPT gained viral attention. AutoGPT was described as an “AI agent” that, when given a goal in natural language, would attempt to achieve it by breaking it down into sub-tasks and executing them autonomously. AutoGPT “wraps” an LLM (like GPT-4) with an iterative loop: it plans actions, executes one, observes the results, and then determines the following action, repeating this cycle until the goal is achieved or the user intervenes. While products like AutoGPT are still experimental and have limitations, they represent a clear move from co-pilot to autopilot, where the user specifies the desired outcome, and the AI endeavors to figure out the methodology.
Specialized Autonomous Agents (e.g., Devin, 2023): More specialized autonomous agents appeared following the general trend. Devin, developed by Cognition Labs, is marketed as an AI software engineer. It can reportedly take a software development task from specification to a functional product, including planning, coding, debugging, and even researching documentation online if it encounters an unfamiliar problem – all with minimal human assistance. This points towards a future where AI agents might specialize in various professional domains.
Multi-Modal and Embodied Agents (Ongoing): Research continues to push AI agents towards interacting with the world in more human-like ways. This includes developing agents that can process and respond to multiple types of input (text, images, sound) and agents that can control physical systems, like robots. Google’s work on models like PaLI-X, which can understand and generate text interleaved with images, and their research into robotic agents that can learn from visual demonstrations, are examples of this trend. The goal is to create agents that can perceive, reason, and act holistically in complex, real-world environments.

If you would like to learn more about AutoGPT, visit my blog post.

AutoGPT: The Game Changer in Artificial Intelligence and Autonomous Agents

Manus AI: A General Agentic AI System

Manus AI is a prominent example of a general-purpose agentic AI system. As described on its website and in various tech reviews, Manus is designed to be “a general AI agent that bridges minds and actions: it doesn’t just think, it delivers results.” It aims to excel at a wide array of tasks in both professional and personal life, functioning autonomously to get things done.

Capabilities and Use Cases (from website and reviews):

Personalized Travel Planning: Manus can create comprehensive travel itineraries and custom handbooks, as demonstrated by its example of planning a trip to Japan.
Educational Content Creation: It can develop engaging educational materials, such as an interactive course on the momentum theorem for middle school educators.
Comparative Analysis: Manus can generate structured comparison tables for products or services, like insurance policies, and provide tailored recommendations.
B2B Supplier Sourcing: It conducts extensive research to identify suitable suppliers based on specific requirements, acting as a dedicated agent for the user.
In-depth Research and Analysis: Manus has been shown to conduct detailed research on various topics, such as AI products in the clothing industry or compiling lists of YC companies.
Data Analysis and Visualization: It can analyze sales data (e.g., from an online store) and provide actionable insights and visualizations.
Custom Visual Aids: Manus can create custom visualizations, like campaign explanation maps for historical events.
Community-Driven Use Cases: The Manus community showcases a variety of applications, including generating EM field charts, creating social guide websites, developing FastAPI courses, producing Anki decks from notes, and building interactive websites (space exploration, quantum computing).

Architecture and Positioning:

While specific deep technical details are often proprietary, reports suggest Manus AI operates as a multi-agent system. This implies it likely combines several AI models, possibly including powerful LLMs like Anthropic’s Claude 3.5 Sonnet (as mentioned in some reviews) or fine-tuned versions of other models, to handle different aspects of a task. This architecture allows for specialization and more robust performance on complex, multi-step projects. Manus positions itself as a highly autonomous agent, aiming to go beyond the capabilities of traditional chatbots by taking initiative and delivering complete solutions.

Check out my blog post if you want more information about Manus AI.

Beyond Chatbots: Understanding Manus AI, the General AI Agent Changing Everything

Nine Cutting-Edge Agentic AI Projects Transforming Tech Today

1. Atera Autopilot (Launching May 20)

What it does: Atera’s Action AI Autopilot is coming to market on May 20, and it will offer users access to a fully autonomous helpdesk AI for IT teams. Our AI Copilot solution has already utilized AI to simplify ticketing and help desk solutions, speeding up ticket resolution times by 10X and reducing IT team workloads by 11-13 hours per week. Autopilot will push the envelope further by taking human agents out of typical help desk situations.

How Autopilot uses Agentic AI: Autopilot leverages Agentic AI to autonomously triage incoming support requests, routing straightforward issues, like password resets or software updates, to self-resolution without human intervention. It also proactively scans system logs for emerging errors, generates and applies fixes in real time, and escalates complex tickets to the right technician only when necessary.

Why it matters: Atera’s Autopilot tool offers large-scale applications for IT service management. Many teams are overwhelmed and understaffed, struggling to deal with demanding support tickets and help desk requests. Autopilot aims to solve this problem with a scalable, user-friendly solution that will improve customer satisfaction and allow IT teams to focus their cognitive skills on more complex, rewarding issues.

2. Claude Code by Anthropic

What it does: Claude Code is an Agentic AI coding tool currently in beta testing. It lives in your terminal, understands your code base, and allows you to code faster than ever through natural language commands. Claude Code, unlike other tools, doesn’t require additional servers or a complex setup.

How Claude Code uses Agentic AI: Claude Code is an Agentic AI experiment that learns your organization’s code base as part of its training data, allowing it to improve over time. You don’t have to add files to your context manually—Claude Code will explore your base as needed.

Why it matters: Coding has been one of the most critical applications of Agentic AI. As these tools grow more advanced, IT teams and developers can take a more hands-off approach to coding, allowing for more efficient and productive teams.

3. Devin by Cognition Labs

What it does: Cognition Labs calls its AI tool Devin “the first AI software engineer.” Devin is meant to be a teammate to supplement the work of IT and software engineering teams. Devin can actively collaborate with other users to complete typical development tasks, reporting real-time progress and accepting feedback.

How Devin uses Agentic AI: Devin uses Agentic AI capabilities through multi-step, goal-oriented pursuits. The program can plan and execute complex engineering tasks requiring thousands of decisions. Devin can recall relevant context at every step, learn over time, and fix mistakes, all requiring Agentic AI.

Why it matters: Devin has already been used in many different real-life scenarios, including helping one developer maintain his open-source code base, building apps end-to-end, and addressing bugs and feature requests in open-source repositories.

4. Personal AI (Personal AI Inc.)

What it does: Personal AI creates AI personas, digital representations of job functions, people, and organizations. These personas work toward defined goals and help complete tasks that human employees might otherwise do.

How Personal AI uses Agentic AI: Each AI persona can make autonomous decisions while processing data and context in real time.

Why it matters: The AI workforce movement, which is embodied in Personal AI, allows you to expand your workforce of real-world individuals without incurring the costs of salaried employees. These AI personas can complement and enhance the work of your human team.

5. MultiOn (Autonomous web assistant by Please)

What it does: MultiOn is an autonomous web assistant created by AI company Please. The tool can help you complete tasks on the web through natural language prompts—think booking airline tickets, browsing the web, and more.

How MultiOn uses Agentic AI: MultiOn completes autonomous actions and multi-step processes following NL prompts.

Why it matters: Parent company Please has emphasized the travel use cases for its Agentic AI bot. However, many scenarios exist where an autonomous web assistant like MultiOn can simplify everyday life.

6. ChatDev (Simulated company powered by AI agents)

What it does: ChatDev is a virtual software company with AI agents. The company is meant to be a user-friendly, customizable, extendable framework based on large language models. It also presents an ideal scenario for studying collective intelligence.

How ChatDev uses Agentic AI: The intelligent agents within ChatDev are working autonomously (both independently and collaboratively) toward a common goal: “revolutionize the digital world through programming.”

Why it matters: ChatDev is an excellent study of Agentic AI’s collaborative potential. It also allows users to create custom software using natural language commands.

7. AgentOps (Operations platform for AI agents)

What it does: AgentOps is a developer platform for building AI agents and large language models (LLMs). It allows companies to develop their Agentic AI workforces through custom agents and then understand their activities and costs through a user-friendly and accessible interface.

How AgentOps uses Agentic AI: The company specializes in building intelligent, Agentic AI agents that can operate autonomously—they can make decisions, take actions, and execute multi-step processes without human intervention.

Why it matters: AgentOps is one of the Agentic AI tools to watch this year. With the growing popularity of AI workforces, building custom agents and tracking them to ensure reliability and performance is set to be a crucial consideration for many organizations.

8. AgentHub (Agentic AI marketplace)

What it does: With AgentHub, you can use easy, drag-and-drop tools to create custom Agentic AI bots. Plenty of workflow templates exist, and you don’t need extensive AI experience to build your personalized AI tools.

How AgentHub uses Agentic AI: While not all AI bots created on AgentHub are Agentic, the bots you can build use more Agentic AI as the features become more advanced.

Why it matters: Tools like AgentHub extend the reach of AI to a broader audience, as you don’t need to be a professional developer or programmer to use and benefit from these frameworks.

9. Superagent (Framework for building/hosting Agentic AI agents)

What it does: Superagent is an AI tool that is focused on creating more and better AI agents that are not constrained by rigid environments. Superagent allows human and AI team members to work together to solve complex problems.

How Superagent uses Agentic AI: Superagent is all about Agentic AI. These agents are meant to learn and grow continuously. They are not restricted by predefined knowledge and are intended to grow with your company rather than quickly becoming obsolete as AI advances.

Why it matters: The Superagent team’s belief system centers around building flexible, autonomous agents, not caged in by fears of AI takeover. Instead, Superagent emphasizes the possibilities for humankind when we work in tandem with AI.

Source: https://www.atera.com/blog/agentic-ai-experiments/

Benefits and Opportunities of Agentic AI

The rise of agentic AI systems brings with it a multitude of benefits and opens up new opportunities across various sectors:

Amplified Productivity: Perhaps the most immediate benefit is a significant boost in productivity. Autonomous agents can work 24/7 without fatigue, handling tedious, repetitive, or time-consuming tasks. This frees human workers to focus on their jobs’ creative, strategic, and interpersonal aspects. For example, a software developer can delegate boilerplate coding to an AI agent, or a researcher can have an agent sift through vast literature.
New Capabilities and Services: Agentic AI enables the creation of entirely new services and makes existing ones more sophisticated. Personalized education tutors that adapt to each student’s learning pace, AI-powered therapy bots (under human supervision) that provide cognitive behavioral exercises, or advanced analytical tools for small businesses that were previously only affordable for large corporations, are becoming feasible.
Accessibility and Empowerment: By encapsulating expertise into an AI agent, specialized knowledge and skills become more accessible to a broader audience. An individual might not be able to afford a team of marketing experts, but an AI marketing agent could help them devise and execute a campaign. Similarly, AI agents could assist with navigating complex legal or financial information (though always with the caveat that they are not substitutes for professional human advice in critical situations).
Continuous Operation and Multitasking: Unlike humans, AI agents don’t need breaks and can handle multiple data streams or tasks in parallel if designed to do so. A customer service operation could deploy AI agents to handle a large volume of inquiries simultaneously, or a security system could use agents to monitor numerous feeds for anomalies around the clock. This continuous operational capability is invaluable in many fields.

Challenges and Risks of Going Autopilot

Despite the immense potential, the increasing autonomy of AI agents also presents significant challenges and risks that must be addressed thoughtfully:

Reliability and Accuracy (Hallucinations): Large Language Models, the core of many agents, are known to sometimes “hallucinate” – producing incorrect, nonsensical, or fabricated information with great confidence. In a co-pilot scenario, a human can often catch these errors. However, if an agent operates autonomously, there’s a higher risk of making a bad decision or producing flawed outputs without immediate human correction. Ensuring reliability is tough and requires techniques like validation steps, cross-referencing, or voting among multiple models, but errors can still occur.
Unpredictable Behavior: When an AI agent is given a broad or vaguely defined goal, it may devise unexpected or undesirable ways to achieve it. The AutoGPT experiment, which reportedly tried to exploit its environment to gain admin access, is one example. Another notorious case was ChaosGPT, an agent prompted with an evil objective (“destroy humanity”), which then researched destructive methods. While these are extreme examples, even with benign intent, an agent might misunderstand a goal or take unconventional, problematic steps.
Alignment and Ethics: A crucial challenge is ensuring that an agent’s actions align with human values, ethical principles, and the user’s explicit (and implicit) instructions. For instance, an AI agent tasked with screening resumes might inadvertently develop biased criteria if not carefully designed, leading to discriminatory outcomes. Embedding ethical guidelines (like Anthropic’s Constitutional AI approach, where the AI is trained with principles to self-check its outputs) and maintaining continuous oversight and robust feedback loops are essential. Regulations may also be needed regarding what autonomous agents can do, especially in sensitive areas like finance or healthcare.
Security Vulnerabilities: Autonomous agents open new avenues for attack. “Prompt injection,” where malicious instructions are hidden within data that an agent processes, can hijack the agent’s behavior. If an agent is connected to many tools and APIs, each connection is a potential point of vulnerability. Ensuring data security and limiting an agent’s permissions (e.g., restricting a file-writing agent to a specific directory) are essential safeguards.
Quality of User Experience: From a practical standpoint, interacting with current AI agents can sometimes be frustrating. They might get stuck in loops, repeatedly fail at a task, or ask for confirmation too frequently for trivial matters. Conversely, they might proceed with a flawed plan if they don’t ask for enough confirmation. Finding the right balance between autonomy and user interaction is an ongoing design challenge.
Job Impact and Social Implications: The potential for AI agents to automate tasks currently performed by humans raises significant concerns about job displacement and the need for workforce re-skilling. While some argue that AI will create new jobs, the transition can be disruptive. There’s also a broader societal impact, such as how the value of human judgment and uniquely human skills might change.
Over-Reliance and Trust: As agents become more competent, there’s a risk that humans may become over-reliant on them or trust their outputs too blindly. This is similar to how people sometimes follow GPS navigation even when it seems to lead them astray. Maintaining a healthy skepticism and understanding the limitations of AI is essential.

The Road Ahead: From Autopilot to… Autonomous Teams?

The journey of agentic AI is still in its early stages. The systems we see today, like AutoGPT or Devin, are pioneering prototypes – sometimes clunky, sometimes astonishing. What might the next few years bring as this technology matures?

Many experts advocate for a gradual approach to autonomy. This means starting with co-pilot systems to build trust and gather data, then slowly introducing more autonomous features in low-risk settings as the kinks are worked out. The goal isn’t necessarily to remove humans from the loop entirely, but to safely expand what humans and AI can accomplish together.

Shortly, we can expect several key developments:

Better Reasoning and Less Hallucination: Intense research focuses on improving how AI models reason and how consistent and factually accurate they are. Techniques like trained reflection (where the AI learns to critique and enhance its own outputs), iterative planning, and incorporating symbolic logic or knowledge graphs alongside LLMs could make agents more reliable. Companies like OpenAI, Google, and Anthropic are explicitly optimizing their models (e.g., future versions of GPT or Gemini) for multi-step tasks and factual accuracy.
Longer Context and Memory: We’ve already seen models like Anthropic’s Claude handle huge context windows (hundreds of thousands of tokens). This trend will continue, meaning agents can remember long dialogues or large knowledge bases during their operations without needing as much external lookup. This reduces the chances of forgetting instructions or repeating mistakes and allows an agent to consider more factors simultaneously.
More Seamless Tool Ecosystems: We’ll likely see tighter and more standardized integrations between AI agents and software APIs. Major software platforms are racing to become “AI-friendly.” We might see standardized “agent APIs” for everyday tasks – a universal way for any AI agent to interface with email, calendars, databases, etc., without custom glue code each time. This would be akin to how USB standardized device connections.
Domain-Specific Autopilots: It’s probable that highly specialized agents, fine-tuned on data and workflows for specific domains (e.g., an “AI Scientist” for drug discovery, an “AI Lawyer” for legal research and document drafting), will outperform general-purpose agents in those niches for some time. These agents will know their limits and when to defer to a human expert, tailored to the workflows of that profession.
Human-Agent Team Structures: As organizations increasingly use AI agents, we’ll likely see new team structures and new roles emerge. A human project manager might coordinate a group of AI agents, each working on subtasks. Conversely, an AI could take on a management role for routine coordination, with humans focusing on creative tasks. Startups like Cognition Labs (behind Devin) have already experimented with an agent that delegates to other agents, hinting at a future where you might launch a swarm of agents for a big goal – an approach sometimes called multi-agent systems. These could collaborate or even compete in a limited way to improve robustness.
Regulation and Standards: With great power comes the need for oversight. We can anticipate regulatory frameworks emerging for autonomous AI, much like we have for self-driving cars. This might include requirements for disclosure (so humans know when they are interacting with an AI), liability frameworks (who is responsible if an AI agent causes harm?), and industry standards or ethical guidelines for AI development and deployment.
Unexpected New Modes of Use: Every time a new AI capability has emerged, users have found creative and surprising ways to use it. Autopilot agents could lead to phenomena we haven’t imagined. One could picture things like highly personalized AI agent companions that know you deeply and help organize your life, or perhaps AI agents representing individuals as proxies in certain situations (e.g., negotiating prices or deals automatically on your behalf within parameters you set). The boundary between “tool” and “partner” will blur as these agents become more present in our daily activities.

Conclusion

The evolution from AI co-pilots to AI autopilots represents a fundamental shift in leveraging machine intelligence. What began as simple assistive tools – helpful but limited – has rapidly advanced into autonomous agents that can handle complex tasks with minimal oversight. We’ve explored how this became possible: the advent of powerful language models, new architectures for memory and planning, and integration with the rich toolsets of the digital world. We’ve also seen concrete examples, from coding assistants that can build entire apps, to business agents scheduling meetings and drafting reports, to experimental agents pushing the frontiers of science and strategy.

The benefits of agentic AI are manifold – increased productivity, the ability to tackle tasks at scale, democratizing expertise, and freeing human potential. Yet, alongside these benefits, we must address challenges: ensuring these agents behave reliably, ethically, and securely; reshaping workflows and job roles thoughtfully; and maintaining human control and trust.

In aviation, autopilot systems have long assisted pilots, but we still rely on skilled pilots to oversee them and handle the unexpected. In a similar vein, AI autopilots will help us in various endeavors, but human judgment, creativity, and responsibility remain irreplaceable. The transition we are experiencing is not about handing everything over to machines but redefining collaboration between humans and AI. We are learning what tasks we can safely delegate to our “digital interns” and where we still need to be firmly in command.

The term “agentic AI” captures the exciting and sometimes unnerving idea of AI that has agency—that can act in the world. As we’ve discussed, we’re already giving AI some agency in controlled ways. In the coming years, we will expand that agency in small steps, test boundaries, and find the right balance of autonomy and oversight. It’s a journey that involves technologists, domain experts, ethicists, and everyday users all playing a part in shaping how these agents are built and used.

From co-pilots that suggest to autopilots that execute, AI systems are becoming more capable and independent. It’s an evolution that promises to profoundly change the nature of work and innovation. Suppose we navigate it wisely – steering when needed, trusting when justified – we could unlock tremendous value while keeping aligned with human goals. Ultimately, the best outcome is not AI running the world on autopilot, nor humans refusing to automate anything; it’s a well-orchestrated partnership where AI agents handle the heavy lifting in the background, and humans steer the overall direction.

In a sense, we are becoming commanders of fleets of intelligent agents. Just as good leaders empower their team but remain accountable, we will empower our AI co-pilots and autopilots, guiding them with a high-level vision and ethical compass. The evolution of agentic AI is the evolution of that partnership. The cockpit has gotten more crowded—we now have AI co-pilots and autopilots joining us—but with clear communication and controls, the journey can be safe and fruitful for all aboard.

That’s it for today!