Getty Images
Agentic AI is being heralded as the future of the generative AI revolution by leaders in the field. From ChatGPT’s integration of agentic features to the rise of Comet (the agent-based web browser from Perplexity) and Chinese-born Manus, the trend of handing more control to AI tools seems inevitable. At least, that’s the view of Microsoft CEO Satya Nadella, Shopify CEO Tobias Lütke, Amazon executive chairman Jeff Bezos, and Nvidia CEO Jensen Huang…….Continue reading….
By: Chris Stokel-Walker
Source: Fast Company
.
Critics:
Agentic AI is a class of artificial intelligence that focuses on autonomous systems that can make decisions and perform tasks without human intervention. The independent systems automatically respond to conditions, to produce process results. The field is closely linked to agentic automation, also known as agent-based process management systems, when applied to process automation.
Applications include software development, customer support, cybersecurity and business intelligence. The core concept of agentic AI is the use of AI agents to perform automated tasks but without human intervention. While robotic process automation (RPA) and AI agents can be programmed to automate specific tasks or support rule-based decisions, the rules are usually fixed. Agentic AI operates independently, making decisions through continuous learning and analysis of external data and complex data sets.
Functioning agents can require various AI techniques, such as natural language processing, machine learning (ML), and computer vision, depending on the environment. Particularly, reinforcement learning (RL) is essential in assisting agentic AI in making self-directed choices by supporting agents in learning best actions through the trial-and-error method. Agents using RL continuously to explore their surroundings will be given rewards or punishment for their actions, which refines their decision-making capability over time.
All the while deep learning, as opposed to rule-based methods, supports agentic AI through multi-layered neural networks to learn features from extensive and complex sets of data. Further, multimodal learning enable AI agents to integrate various types of information, such as text, images, audio and video. As a result, agentic AI systems are capable of making independent decisions, interacting with their environment and optimising processes without a human directly intervening.
AI agents can be used to perform small tedious tasks during web browsing and potentially even perform browser actions on behalf of the user. Products like OpenAI Operator, Perplexity Comet and Dia (from The Browser Company) integrate a spectrum of AI capabilities including the ability to browse the web, interact with websites and perform actions on behalf of the user. In 2025, Microsoft launched NLWeb, a agentic web search replacement that would allow websites to use agents to query content from websites by using RSS-like interfaces that allow for the lookup and semantic retrieval of content.
Products integrating agentic web capabilities have been criticised for exfiltrating information about their users to third-party servers and exposing security issues since the way the agents communicate often occur through non-standard protocols. In 2025, MIT’s study revealed that about 95% of enterprise generative-AI pilots fail to deliver measurable P&L impact[22]; a failure rate in business outcomes.
The report titled “The GenAI Divide: State of AI in Business 2025,” based on 150 executive interviews, a survey of 350 employees, and analysis of 300 deployments, and it attributes the failures largely to integration issues. The MIT NANDA report finds only about 5% of corporate generative-AI pilots are achieving rapid revenue acceleration, with the vast majority showing little to no impact on profit and loss statements.
It also notes a mismatch in spending (heavy on sales/marketing tools versus higher ROI in back-office automation) and highlights “shadow AI” usage complicating measurement and governance. One of the first reason on study was about “integration gap”, that chatbot does not find enough time to deploy themselves and adopt the workflow environment, leading to little to no measurable impact. Over half of budgets go to sales/marketing tools, while bigger returns often lie in back-office automation that reduces outsourcing and agency costs.
Study also notes that while building internal tools is possible, parternership with external agency is often more profitable than not. The study explicitly reports that external partnerships have about twice the success rate of internal builds (~67% vs ~33%), often yielding faster time-to-value and lower total cost, so buying/partnering can be more profitable than building in-house given today’s high failure rates.
With roughly 95% of enterprise GenAI efforts failing to reach measurable P&L impact, failed internal tools represent a significant sunk cost risk compared with proven vendor solutions that integrate and learn within workflows. For example, building your own AI tool that may not work as expected can result in loss than to buy an externel tool that perform the tasks more easily, resulting in profit.
Leave a Reply