The Era of Agentic AI!
AI agents will transform the way we interact with technology, making it more natural and intuitive. They will enable us to have more meaningful and productive interactions with computers.
This quote, from the godmother of AI, Fei-Fei Li, well explains what’s happening next in AI!
Fei-Fei Li was the computer scientist who pioneered the field with ImageNet, the first breakthrough which made clear that AI was here to stay, and that opened up the way to what would become modern transformers.
At this moment she’s working on spatial intelligence, which will be a key domain for the development of real-world applications for AI!
We take for granted that computers are tools to which we need to explicitly program what we want to achieve before they can help us solve very specific, narrow tasks. Yet, starting in 2022, this has changed.
The whole computing paradigm is now completely reshaped to make computers the tools we speak to. The new generations will take this for granted. Computers aren’t boxes to which you type in, these are agents to which you can talk, about anything!
That’s the era of Agentic AI, and we’re right in the middle of it!
Agentic AI refers to artificial intelligence systems capable of autonomous action and decision-making.
These systems, often called AI agents, can pursue goals independently, make decisions, handle complex situations, and adapt to changing environments without direct human intervention.
They leverage advanced techniques such as reinforcement learning and evolutionary algorithms to optimize their behavior and achieve specific objectives set by their human creators.
Keep in mind that there is no single definition of Agentic AI.
Agentic AI in academic settings might be more about "agency" or the ability of these AI agents to make complex decisions independently.
In business, for the next couple of years, agentic AI will primarily concern specific business outcomes and tasks that these agents can achieve in a very constrained environment to ensure their accuracy, reliability, and security are a priority.
Read Next
What made Agent AI something else compared to the initial wave of AI?
Since the launch of GPT-2 in 2019, the Gen AI paradigm has been based on prompting - for the last five years.
In short, the LLM completed any task based on a given instruction. The quality of the output highly depended upon the quality of the input (prompt).
However, in the last few weeks, we’ve finally seen the rise of Agentic AI, a new type of artificial intelligence that can solve complex problems independently using advanced reasoning and planning.
Unlike regular AI, which responds to single requests, agentic AI can handle multi-step tasks like improving supply chains, finding cybersecurity risks, or helping doctors with paperwork.
It works by gathering data, devising solutions, carrying out tasks, and learning from the results to improve over time.
What are the critical features of Agentic AI vs. Prompting?
• Autonomous Problem-Solving: Agentic AI uses sophisticated reasoning and iterative planning to solve complex, multi-step tasks independently.
• Four-Step Process: Perceive (gathers data), Reason (generates solutions), Act (executes tasks via APIs), and Learn (continuously improve through feedback).
• Enhanced Productivity: Automates routine tasks, allowing professionals to focus on more complex challenges, improving efficiency.
• Data Integration: This technique uses techniques like Retrieval-Augmented Generation (RAG) to access a wide range of data for accurate outputs and continuous improvement.
When did this Agentic AI wave start?
It all started two years back.
Indeed, Chain-of-Thought Prompting (CoT), the paper that spurred the current Agentic AI wave, only came out in early 2022.
Like the Transformer paper (Attention Is All You Need), which came out in 2017 and led to ChatGPT, CoT was also an effort of the Google Research and Brain Teams!
The "Chain-of-Thought Prompting" (CoT) paper, published in early 2022 by researchers from Google's Research and Brain Teams, has been pivotal in advancing the capabilities of large language models (LLMs).
This technique enhances LLMs' reasoning abilities by guiding them to generate intermediate steps that mirror human problem-solving processes by:
Enhanced Reasoning Capabilities: CoT prompting enables LLMs to tackle complex tasks by breaking them down into sequential steps, improving performance in areas like arithmetic, commonsense reasoning, and symbolic manipulation.
Emergent Abilities with Scale: The research demonstrated that as LLMs increase in size, their capacity for chain-of-thought reasoning naturally emerges, allowing them to handle more intricate problems effectively.
Influence on Agentic AI Development: The CoT paper has inspired the development of agentic AI systems capable of more autonomous and sophisticated decision-making by showcasing how LLMs can perform complex reasoning through structured prompting.
This progression mirrors the impact of the 2017 "Attention Is All You Need" paper, which introduced the transformer architecture and laid the groundwork for models like ChatGPT.
Both papers underscore the significant role of Google's research teams in propelling advancements in AI, particularly in enhancing the reasoning and comprehension abilities of language models.
And yet, guess what?
Most of them pushed OpenAI forward from a commercial application standpoint. OpenAI's GPT-4o, released in May 2024, incorporates principles from CoT to improve its reasoning capabilities.
By structuring prompts to encourage step-by-step thinking, GPT-4o can more effectively handle complex tasks such as mathematical problem-solving and logical reasoning.
This approach allows the model to break down intricate problems into manageable steps, leading to more accurate and coherent responses.
The race has heated up so much that, as rumors came out of OpenAI Orion, a next-generation AI model developed by OpenAI, became a massive hit!
OpenAI Orion is the rumored next-generation AI model developed by OpenAI, designed to significantly enhance reasoning, language processing, and multimodal capabilities.
It is expected to be 100 times more potent than GPT-4, with the ability to handle text, images, and videos seamlessly.
Initially intended for key partner companies and not for broad public release, Orion aims to revolutionize various industries by providing advanced problem-solving and natural language understanding capabilities.
Thus, it advances OpenAI's vision towards artificial general intelligence (AGI) and strategic collaborations with Microsoft Azure.
And OpenAI is not alone there!
After rumors a few weeks back, it seems that Google has actually, even if briefly, leaked an AI prototype, “Jarvis,” designed to complete computer tasks like booking flights or shopping.
Despite being available temporarily on the Chrome extension store, the tool didn’t fully work and was quickly removed. Google planned to unveil Jarvis in December, joining competitors like Anthropic and OpenAI in AI assistance.
What happened there? As reported by The Information:
Accidental Release: Google briefly publicized an internal AI prototype, codenamed “Jarvis,” designed to complete tasks on a person’s computer.
Capabilities: Jarvis, a “computer-using agent,” aims to assist with tasks like purchasing products or booking flights.
Access Issue: The prototype, available via the Chrome extension store, didn’t function fully due to permission restrictions.
Removal: Google removed the product by midafternoon; it was intended for a December release alongside a new language model.
Competition: Anthropic and OpenAI are also developing similar AI task-assistance products.
What can we expect there?
Agentic AI: Personal, Persona-Based, Company Agents
The Academic Definition focuses on AI Agents as systems that reason and act autonomously, originating from the concept of "agency."
I love the business definition, which the CEO of Sierra Bret Taylor, gave on the podcast No Priors in episode number 82, where he explained there are, according to him, three main kinds of agents we'll see emerge there:
Personal Agents: Help individuals with tasks like managing calendars or triaging emails.
Persona-Based Agents: Specialized tools for specific jobs (e.g., coding or legal work).
Company Agents: Customer-facing AI that enables businesses to engage digitally with their users.
More precisely:
Here’s the breakdown of the three types of agents, along with potential business models for each:
Personal Agents
Agents assist individuals with tasks like managing calendars, triaging emails, scheduling vacations, and preparing for meetings.
State of Development: Early-stage; complex due to broad reasoning requirements and extensive systems integrations.
Challenges: High complexity in task diversity and integration with personal tools.
Potential Business Models:
Subscription-Based Services: Charge users a recurring fee for access to personal assistant functionalities (e.g., premium tiers for advanced features).
Freemium Models: Offer basic features for free, with paid upgrades for advanced integrations and additional automation.
B2B Partnerships: Collaborate with productivity tool providers (e.g., Google Workspace, Microsoft 365) to integrate and sell personalized solutions.
Licensing: License the technology to companies creating proprietary productivity tools or devices (e.g., smartwatches, phones).
Persona-Based Agents
Specialized agents tailored for specific professions or tasks, such as legal assistants, coding assistants, or medical advisors.
State of Development: Mature in certain niches with narrow but deep task scopes.
Examples: Harvey for legal functions and coding agents for software development.
Advantages: Focused engineering and benchmarks streamline development.
Potential Business Models:
Vertical SaaS (Software as a Service): Offer domain-specific AI tools as subscription-based services targeted at professionals (e.g., lawyers, developers).
Pay-Per-Use: Monetize by charging based on usage or the number of completed tasks.
Enterprise Licensing: Provide customized agents for large organizations in specific industries.
Marketplace Integration: Integrate with platforms like GitHub (for coding agents) or Clio (for legal agents) and earn through platform fees or partnerships.
Company Agents
Customer-facing agents represent companies, enabling tasks like product inquiries, commerce, and customer service.
State of Development: Ready for deployment with current conversational AI technology.
Vision: Essential for digital presence by 2025, akin to having a website in 1995.
Potential Business Models:
B2B SaaS: Offer branded AI agents as a service to companies, providing monthly or annual subscription plans based on features and scale.
Performance-Based Pricing: Charge companies based on metrics like customer satisfaction, retention rates, or reduced operational costs.
White-Label Solutions: Provide customizable AI agent templates that companies can brand as their own.
Integration Fees: Earn from integrating AI agents into companies’ existing CRM, e-commerce, or support systems.
Revenue Sharing: For commerce-related interactions, take a small percentage of sales the AI agent facilitates.
Sierra Bret Taylor's CEO also emphasized that we'll see these agents evolve at hardware and software levels.
What will be the next device to enable AI Agents?
The Smartphone will be the "Central Hub of AI" in the initial phase
While, over time, AI might enable a whole new hardware paradigm and form factor, it's worth remembering that the first step of integrating AI is happening within the existing smartphone ecosystem.
In short, the smartphone will remain the "Central Hub of AI" in the next few years until a new native form factor evolves.
However, in the next 3-5 years, the iPhone will remain a key platform for the initial development of AI.
Take the AI iPhone (trend data below):
In Apple's latest iPhone models, specifically the iPhone 16, Apple integrated advanced artificial intelligence (AI) capabilities known as "Apple Intelligence" (trend data below):
Apple Intelligence is a suite of generative AI capabilities developed by Apple, integrated across its products including iPhone, Mac, and iPad.
Still, this system will enhance features like Siri, writing, image creation, and personal assistant functionalities at the embryonal stages.
It aims to simplify and accelerate everyday tasks while prioritizing user privacy through on-device processing and Private Cloud computing.
In the meantime, the smartphone will be the first device to be completely revamped before we see the emergence of AI-native devices, such as combining AR with them.
For now, on the smartphone side, the AI revolution in smartphones is moving toward hyper-personalization, with each player giving it its own twist:
Apple champions privacy with on-device AI,
Samsung boosts performance through smart optimizations,
Google elevates photography with stunning enhancements,
Huawei adds practical tools for everyday ease.
Each brand brings unique AI-driven features, turning phones into powerful personal assistants.
Below is the breakdown of each smartphone player's AI strategy:
Apple iPhone: Apple focuses on blending privacy with advanced AI capabilities. With its Apple Intelligence platform, the iPhone offers tools like a language model for email and document management and creative features like Image Playground and Genmoji. Apple’s strong commitment to on-device processing minimizes data transmission, appealing to privacy-conscious users.
Samsung Galaxy: Samsung’s Galaxy S24 Ultra, featuring the Exynos chipset, emphasizes high performance with AI-optimized cores. Its Scene Optimizer camera feature automatically adjusts settings for various scenes, while intelligent performance optimization enhances responsiveness and extends battery life, making it a robust option for power users.
Google Pixel: Known for its photography, its Tensor chip powers features like Magic Eraser for object removal in photos, AI-enhanced zoom, and low-light photography. The Gemini chatbot enhances communication, providing real-time captions, transcription, and translation, positioning Pixel as the top choice for photo and communication aficionados.
Huawei Pura 70 Series: Huawei’s AI focuses on practical enhancements. With features like Image Expand for background filling, Sound Repair for call quality, and an upgraded Celia assistant for image recognition, Huawei offers real-world AI solutions for daily convenience.
Re-emergence of Smart Assistants to reduce screen time?
As I'll show you further down the research, as we're closing 2024, Apple, Google, and Amazon are all "secretly" working on revamping their smart assistants.
The wave started a decade ago when these big tech players tried to dominate the "voice assistant" market and ended up as a missed promise.
These assistants are not delivering on their promises. Take Siri, which turned out to be a long-term flop because of its lack of usefulness.
Yet, will we see the renaissance of these devices via Generative AI?
For instance, smart speakers (e.g., Alexa, Siri, Google Home) and headphones may become central to daily workflows.
Conversational interfaces in these devices could enable seamless, screen-free engagement for tasks like scheduling, reminders, or information retrieval.
Beyond the Smartphone form factor
Yet, we might figure out new form factors in the coming decade. Indeed, while the smartphone will remain the primary computing device for most users, how we interact with it is evolving.
Conversational AI and multimodal interfaces will integrate seamlessly into everyday experiences, reducing our dependence on screens.
Evolution of Customer Experiences
As Sierra Bret Taylor's CEO also emphasized, we might see these interesting trends with AI agents:
From Menus to Conversations:
The shift from rigid menu-driven systems to free-form conversational agents represents a significant evolution in customer interaction. Users can directly articulate their needs in natural language instead of navigating through predefined paths (e.g., website categories or phone menus). AI will process and act on these requests instantly.Agents as Digital Front Doors:
Just as websites became a company's digital front door in the 1990s, conversational AI agents will become the primary mode of engagement by 2025. These agents will handle customer service inquiries and eventually manage all interactions with businesses, such as product browsing, transactions, and post-sales support.Hyper-Personalized Interactions:
AI agents will offer tailored experiences, adjusting their tone, content, and functionality based on user preferences and history. For instance, an AI agent for a luxury brand might adopt a more formal and polished tone, while one for a casual retailer might use friendly, conversational language.Customer-Centric Ecosystems:
The real-time nature of conversational agents allows businesses to be more agile in responding to customer needs. For instance, if a retailer introduces a new product, an AI agent can instantly acquire the necessary knowledge and incorporate it into interactions—something that would take weeks to implement in a traditional call center.
The Emergence of A New Business Model Paradigm?
Agentic AI systems are poised to reshape industries and create new revenue opportunities. Let’s explore some of the business models this Agentic AI wave might prompt!
1. Personal Agents
Purpose: Assist individuals with everyday tasks, such as managing schedules, emails, or travel planning.
Business Models:
Freemium Model:
Basic functionalities (e.g., calendar syncing, task reminders) offered for free.
Advanced features (e.g., multi-system integration, personalized analytics) unlocked via subscription.
B2B Licensing:
Partner with productivity tool providers like Microsoft 365 or Google Workspace to integrate and license personal agent features.
Subscription Services:
Charge individuals a monthly/annual fee for premium assistant functionalities, such as advanced scheduling, document summarization, and real-time updates.
Integrated Hardware Solutions:
Pair AI personal agents with smart devices (e.g., smartwatches, phones) and bundle them into device offerings.
2. Persona-Based Agents
Purpose: Serve niche professions (e.g., legal, coding, healthcare) by offering domain-specific expertise.
Business Models:
Vertical SaaS:
Offer specialized AI tools tailored for specific industries, like coding agents for developers or diagnostic agents for doctors, on a subscription basis.
Usage-Based Pricing:
Monetize based on the number of tasks or amount of data processed. For example, charge per legal case review or coding session.
Enterprise Licensing:
Provide large organizations with tailored AI agents to streamline workflows, such as contract analysis for legal teams or error detection for coding teams.
Marketplace Integration:
Embed agents into industry-specific platforms (e.g., GitHub for coding or Clio for legal practice management) and earn through integration fees or shared revenue.
3. Company Agents
Purpose: Represent businesses in customer interactions (e.g., support, commerce, and engagement).
Business Models:
B2B SaaS:
Charge businesses a subscription fee to deploy branded AI agents across their customer engagement platforms (websites, apps, chatbots).
Performance-Based Pricing:
Align pricing with outcomes, such as customer satisfaction scores, retention rates, or sales conversions.
Revenue Sharing:
For commerce-focused agents, take a percentage of the revenue generated through AI-managed interactions or transactions.
White-Label Solutions:
Offer customizable AI agents businesses can brand and tailor to fit their specific needs.
Integration Fees:
Charge for seamless integration of AI agents into existing CRM or e-commerce platforms.
4. Industry-Specific AI Agents
Purpose: Transform key sectors like healthcare, education, and finance with intelligent automation.
Business Models:
Healthcare:
Subscription or usage-based pricing for patient diagnostic agents, appointment scheduling bots, or medical coding assistants.
Education:
Licensing AI tutors or curriculum design agents to educational institutions or platforms.
Finance:
Premium services for portfolio management, fraud detection, or regulatory compliance.
5. Agent Development and Customization
Purpose: Enable organizations to build and deploy their own agentic AI systems.
Business Models:
Platform Licensing:
Provide businesses with an Agentic AI platform for building their own agents, charging based on the number of agents deployed or complexity of integrations.
Consultation Services:
Offer expert guidance to organizations looking to design and implement customized AI agents.
APIs and SDKs:
Monetize through APIs or SDKs that developers use to integrate Agentic AI into their existing systems.
6. AI Infrastructure and Ecosystem Support
Purpose: Offer the backend tools and frameworks for deploying Agentic AI.
Business Models:
Infrastructure as a Service (IaaS):
Provide scalable cloud-based solutions for AI agent deployment, offering storage, computation, and training environments.
Developer Tools:
Sell tools for testing, training, and refining AI agents, such as debugging environments and data optimization platforms.
Recap: In This Issue!
Agentic AI represents a transformative paradigm in artificial intelligence, where systems act autonomously to solve complex problems through reasoning and planning.
These AI agents can gather data, analyze it, devise solutions, execute tasks, and learn iteratively, optimizing their performance over time.
Key Features
Autonomous Problem-Solving: Ability to handle multi-step, complex tasks independently.
Four-Step Process: Perceive, Reason, Act, and Learn, enabling continuous improvement.
Enhanced Productivity: Automates routine tasks, freeing up professionals for strategic challenges.
Data Integration: Leverages Retrieval-Augmented Generation (RAG) to ensure accurate and context-rich outputs.
Historical Milestones
The 2017 "Attention Is All You Need" paper introduced the transformer architecture, leading to models like ChatGPT.
The 2022 "Chain-of-Thought Prompting" (CoT) paper enhanced reasoning in AI models by structuring tasks into logical steps, laying the groundwork for Agentic AI.
Recent Developments
OpenAI's GPT-4o and rumors of OpenAI Orion demonstrate the rapid progress in AI capabilities, focusing on reasoning, multimodal processing, and efficiency.
Companies like Google (Jarvis prototype) and Anthropic are also racing to create advanced task-assistance AI systems.
AI Agent Types and Applications
Personal Agents: Manage personal tasks (e.g., calendars, emails).
Business Model: Freemium services, B2B partnerships.
Persona-Based Agents: Tailored for specific professions (e.g., legal, coding).
Business Model: Subscription-based SaaS, enterprise licensing.
Company Agents: Customer-facing solutions for digital engagement.
Business Model: Performance-based pricing, white-label solutions.
Evolving Customer Experiences
Transition from menu-based interfaces to conversational agents, offering seamless natural language interactions.
By 2025, AI agents will serve as digital front doors for businesses, managing end-to-end customer engagements.
Hyper-personalized interactions will adapt tone and content to user preferences, enhancing user satisfaction.
Business Models
1. Personal Agents
Purpose: Assist individuals with personal tasks like managing calendars, emails, or vacation planning.
Business Models:
Freemium: Free basic features; advanced capabilities for a subscription fee.
Subscription Services: Monthly or annual charges for personalized assistant functionalities.
B2B Partnerships: Integration with productivity suites like Google Workspace or Microsoft 365.
Licensing: Offer the technology to third-party developers for proprietary tools.
2. Persona-Based Agents
Purpose: Specialized tools tailored for professions like legal advisors, coders, or healthcare assistants.
Business Models:
Vertical SaaS (Software as a Service): Domain-specific subscriptions targeting professionals.
Pay-Per-Use: Monetization based on completed tasks or time used.
Enterprise Licensing: Customized agents for large organizations with specific industry needs.
Marketplace Integration: Partnering with industry platforms like GitHub or Clio for revenue sharing.
3. Company Agents
Purpose: Customer-facing agents managing interactions, product inquiries, and e-commerce tasks.
Business Models:
B2B SaaS: Subscription-based pricing for branded AI agents tailored to business needs.
Performance-Based Pricing: Fees tied to metrics like customer satisfaction or reduced costs.
White-Label Solutions: Offer customizable agent templates for company branding.
Integration Fees: Revenue from embedding agents into CRM or e-commerce systems.
Revenue Sharing: Percentage of sales facilitated by the AI agent
Future Hardware Trends
Smartphones as AI hubs: Current focus is on integrating AI into smartphones, enabling hyper-personalized, privacy-centric, and performance-driven features.
Potential emergence of AI-native devices combining AR and conversational interfaces for immersive experiences.
Read Next
Ciao!
With ♥️ Gennaro, FourWeekMBA
Could you share some real-life examples of how Agents are used and what the results have been so far?