Agentic: AI that Can Do Things
Agentic AI, this year’s hottest new AI term, isn’t just more innovative software — it’s AI that can act for you and with you. The real question isn’t what it can do, but whether you (and your organization) are prepared to use it. I want to warn the buyer (“Caveat Emptor”). Agentic AI is not for everyone. In fact, it is for very few people (the same individuals in your IT department who build long-lasting products and services).
In this post, I will share:
- A definition, ala McKinsey, of the term (including lower and higher complexity agents)
- My perspective: Be cautious — agents aren’t suitable for everyone
- Six tips for implementing agents
Agentic AI: Formal Definition
The Sept-2025 McKinsey report, “Reimagining life science enterprises with agentic AI” [1], defines Agentic as:
“AI agents are goal-driven systems that operate independently by breaking down complex tasks, interacting with other systems, and learning in real time. They use machine learning and rules-based AI to enable reasoning, memory, and the capacity to interact with humans.
Gen AI includes lower-complexity agents, sometimes referred to as “low-code” or “no-code,” that employees with minimal coding experience can create and modify using natural language on various platforms. It also includes higher-complexity agents, sometimes referred to as “procode,” which must be developed and fine-tuned by data scientists or engineers.”
The promise is significant — Agentic AI is more than just search or chat — agents can actually perform tasks.
Based on this McKinsey definition, it is clear that higher-complexity agents require the full IT development cycle, including clear goal setting, infrastructure planning, dev-sec-ops, and more. Add in the evolving maturity levels of the foundational model, and you’re looking at development with changing tools and infrastructure. Worst-case scenario: planned failure; best-case: a costly nightmare.
But what is the potential of lower-complexity agents?
Let me start my answer with a question: how many users in your organization know how to use:
- VLOOKUP in Excel
- A rule in email
- Automatic Table of Contents
- Zapier or Make to automate
This is just the beginning. Even for a user with such digital skills, developing agents that deal with the non-deterministic behavior of current AI will be challenging
In short, using low-code, simple agents requires much more expertise.
The bottom line is simple: there is no such thing as low-code/fast/straightforward agents. For an AI agent to be meaningful, in the short and medium term, you need to follow the whole process of any IT/Development process.
Turn Hype into Reality – Yesha’s Recommendation:
- Identify a clear business case where AI agents (also known as automation) can help.
- Design it carefully, including several variations.
- Ensure you have the proper infrastructure (access to the foundational model, Active Directory, and data sources).
- Be prepared to measure usage and results, and to adjust your plans.
- Implement centrally with sufficient users.
- Be ready to scale (if successful—and consider costs), or pivot or retire.
Takeaway
Agentic AI promises a new era, but realizing that promise requires maturity — in skills, systems, and mindset. The key is to stay grounded: design for clarity, measure impact, and grow responsibly.
While Agentic AI may be new, the traditional aspects of digital transformation — technology, business, culture, and leadership — still need to be managed.
More Information
- [1] McKisey’s extensive report on Agentic AI in one industry – pharma – Reimagining life science enterprises with agentic AI