|
| Agentic AI This week's newsletter takes a closer look at a shift that's dominating the news right now: Agentic AI.
If you're confused about what Agentic AI is, you're not alone. Check out the Foundations section below for a primer and what Agentic AI is and what it isn't.
This is both a real transition to more autonomous AI activity, and a commonly misused buzzword. What we're talking about in this newsletter is AI moving beyond a chatbot that respond to requests into systems that can act (somewhat) independently to pursue goals, make decisions, and coordinate with other AI agents.
Anthropic's Claude Cowork shows what this looks like in practice: instead of asking a chatbot questions, you give the AI a task, and it works alongside you on shared documents on your computer.
This is a transition from "human-in-the-loop" to "human-governed" systems, where people set guardrails rather than manage every step. However, most organizations are behind the curve in understanding when, where, and how to create the necessary governance and guardrails for safety and success. | Agent-action loops replace user-learning loops. AI systems improve efficiency through interaction with users and other systems, generating information and value autonomously. | AI gets proactive with Claude Cowork, shifting the interface from chatbots to a shared workspace where both human and AI work on the same documents simultaneously. | Many manual tasks can now be automated as retail merchants use agent "squads" that cleanse data, run pricing simulations, and flag actions in unified reports. |
|
| The rise of agentic AI will require companies to rethink their digital business models. For two decades, platforms operated on user-learning loops, where human activity generated the data to improve algorithms. Netflix recommendations got better as you watched. Amazon refined suggestions based on purchases.
The shift to agentic AI changes this dynamic entirely. Rather than passively waiting for user input, AI agents now initiate actions and pursue goals independently. The business model evolution isn't about better learning, it's about AI systems that can take initiative without constant human intervention. People shift from operating the system to setting guardrails and defining objectives while agents determine execution. Strategic Insight: The most significant development is "agent-to-agent" interaction, where AI agents exchange information and coordinate to achieve their goals. Rather than departments using AI independently, now they're sharing context, insights and aligning activities. |
Anthropic's Claude Cowork brings AI directly into your local file system. Users designate a folder on their computer where Claude can read, edit, and create files without uploading to the cloud. Rather than asking questions and receiving answers, users and AI collaborate on actual documents, spreadsheets, and project files stored locally. The system maintains context across multiple files and tasks, enabling iterative refinement that mirrors how humans actually work on complex projects. The Practical Angle: This shift from AI suggesting edits to AI making edits requires a fundamental change in trust and access control. When Claude can actually create, modify, or delete your files, the question becomes, "Are you ready to grant AI direct access to your files?" View Article → McKinsey's retail merchandising study reveals a stark gap between AI promise and reality. While 71% of merchants report that AI tools have had limited impact and 61% say their organizations aren't prepared to scale AI. Conversely, organizations that properly implement AI agents are seeing significant revenue and margin lifts. AI agents are automating up to 60% of manual tasks and flagging priority decisions in unified dashboards, shifting merchants from reactive firefighting to proactive decision approval. Strategic Takeaway: Agentic AI deployment is powerful but must be paired with clear guardrails on pricing thresholds, promotion rules, and cross-category strategies. Without these, retailers risk inconsistencies that customers quickly notice which threatens loyalty. View Article → | Quick HitsGuide to AI Agents in 2026 Following Meta's Manus AI acquisition, the market exploded. The agentic era has three tiers: specialized agents for single tasks, orchestration agents managing workflows, and General Utility Agents capable of broad reasoning across multiple systems. | Agentic AI Security As AI agents gain autonomy, their vulnerabilities become critical risks. Cisco's "semantic firewalls" inspect meaning and intent to prevent prompt injection attacks. This will be a huge growth area for years to come as we discover more attack vectors against AI agents. | Foundations What Really Is Agentic AI? Not all "agents" are actually autonomous, and those that are may not be fully autonomous. Autonomy is a spectrum, and "agentic" has become a buzzword that doesn't always reflect capabilities. True agentic AI can plan multi-step tasks, choose which tools to use, and adapt when something goes wrong. But even these tools are designed with limits, as they should be. Understanding the strengths, weaknesses, and boundaries of any AI agent is key to making technology decisions based on operational reality rather than marketing hype. |
|
| Industry DevelopmentsHome Depot and Google Launch Agentic AI Home Depot's partnership with Google moves AI agents into brick-and-mortar retail. Customer-facing agents manage project lifecycles like building a deck, while internal agents use vision to direct employees to priority tasks, showing how agents can augment rather than replace frontline workers. | Government Backs AI Agents for Heart Disease Care ARPA-H's CLINIC initiative deploys agents for heart disease care to schedule appointments, adjust medications, and order tests, but only after proving they're as safe as human clinicians in simulation. High-stakes healthcare requires rigorous validation before agents enter clinical workflows. |
|
| |
|
