Agentic AI is already creating real value. Organizations that are leading and winning with AI aren’t waiting for perfect conditions. They’re iterating quickly at a small scale, leveraging automation and off-the-shelf tools whenever possible, and getting started with trusted subsets of their data.

Copilots and assistants are now embedded in software of all kinds as general-purpose productivity aids, delivering real gains. Sales teams create better backgrounders, marketing teams generate ideas faster, and engineers code faster. If you’re not using AI daily, you’re shoveling cash into a bonfire. 

[CTA_MODULE]

Why you need AI: Gaining from productivity zero

AI is an unmatched tool for retrieving and synthesizing information. What’s really exciting, though, is not the promise of growing your personal productivity by 10% or 15%, but unlocking new capabilities that didn’t exist before. I call this gaining from productivity zero.

For example, 3 years ago, I’d never used Python or built a Streamlit application. I quite literally started at 0% productivity. With the help of generative AI, I’ve since built over 100 Streamlit applications. Each one solves a specific business problem, such as QBR generators for sales teams, dynamic pricing apps for retail, and freshman retention analyzers for higher ed institutions.

In other words, AI makes entirely new forms of creative and intellectual work accessible to people with good ideas and pressing needs, but limited implementation-level chops (or the time to learn them). It is a power tool for the human mind. 

You can think of the continuing progression from generative to true agentic AI as follows:

1. Pure generative, input/output models, think ChatGPT 3.5 and its immediate predecessors.
2. Tool-augmented LLMs with light agentic capabilities, such as current (early 2026) iterations of ChatGPT, Gemini, Claude, etc.
3. Workflow agents, which are models using APIs and UI automation to automate short sequences of actions
4. Long-horizon agents that are true virtual coworkers that can sustain work over long timeframes, maintain context, and self-correct (or ask for support)

Start simple to build AI agents that work

Agentic AI combines the information retrieval and synthesis capabilities of generative AI with automation through computer use, creating opportunities to gain from productivity zero in every industry.

To get started, you’ll first need:

1. A specific impactful business use case you can actually solve that is fit for purpose for generative or agentic AI
2. A subset of your data, perhaps two or three sources, that can support that use case
3. An architecture/platform approach that lets you deliver value quickly. Data movement, storage, compute, and AI are all easier with platforms rather than custom builds

1. Identify the right use case

Good use cases for generative AI and agentic AI tend to involve:

1. High-volume, repetitive workflows that follow discernible patterns and consume human time
2. Comprehension and synthesis of language (including code)
3. Clear, easily verified success criteria
4. Good data access and context boundaries, such that you can easily find the right data to support the use case, and there are clear standards for appropriateness, privacy, compliance, etc.
5. Natural decomposition into steps that are easily followed and audited, such as computer tool use

I built a healthcare agent to monitor clinical data for safety events and quality indicators. This agent called data from several tools including a patient analyst, a clinical care data analyst, a patient risk assessor, search services, an ROI calculator, and a risk trajectory analyzer. Then the agent synthesized alerts and reports as needed.

It’s tempting to ask a single agent to do everything, but focused, individual agents attached to a limited set of specialized tools seem to work much better. I have found that your context or instruction set can be fairly extensive, provided the scope of the agent’s work is limited.

Agentic AI is not yet at the point where it can bootstrap its own learning or operate with minimal supervision. In addition, agent-to-agent communication and orchestration are still maturing, so current agents still require considerable human supervision and management.

2. Identify and integrate the data to support your use case

The greatest value from AI comes from enriching it with your unique, private, enterprise data. Otherwise, AI can only provide generic outputs, with no ability to tailor them to the unique circumstances of your organization.

I recommend identifying two or three data sources relevant to your use case that you understand well and trust. 

Automated data integration and data movement with tools like Fivetran eliminates the need to spend engineering time designing, building, testing, and deploying custom end-to-end data pipelines. Instead, you can use automated data integration to set and forget data connections.

3. Set up your supporting architecture

Platforms give you simplicity and off-the-shelf capabilities you can’t get anywhere else. You avoid dealing with architectural complexity and get reliability, scalability, and sustainability built in.

No individual company in retail, manufacturing, healthcare, transportation, or any industry can build complementary tools like BigQuery, Google Cloud Storage, Vertex AI, and Gemini better than Google can. Thousands of customers use those platforms, and Google knows how to do it at a trusted scale.

Once you have a good use case, data integration, and platform-based tools, you can focus on what matters to your business: value-impacting applications. It should take only days (or hours) to familiarize yourself with the tools and deliver a working prototype to demonstrate value before scaling.

Some tips:

1. Start incrementally and don’t try to solve everything at once
2. Measure downstream real business outcomes like time saved, costs reduced, or decisions improved
3. Move fast and iterate because waiting for the ideal time means you’re already behind

Examples of agents

I have previously discussed an example involving ranching and a smart weather alert system. The following two examples come from healthcare:

Hospital supply chain performance monitor

This agent continuously monitors comprehensive hospital supply chain data across multiple facilities to analyze inventory health trends, vendor performance correlations, and stockout risk patterns for supply chain officers and procurement teams seeking real-time medical inventory insights. It automatically performs the following: 

• Analyze overall supply chain health performance across all hospital facilities
• Direct immediate attention to inventory items with declining stock levels
• Generate comprehensive stockout risk reports for procurement review
• Identify patient characteristics that impact medical supply usage
• Analyze vendor performance consistency and identify reliability gaps

Healthcare supply chain financial analyzer

Similarly, this agent analyzes and monitors the financials of a hospital supply chain:

• Calculate ROI for hospital inventory management investments across facilities
• Identify inventory cost reduction opportunities through preventive management
• Generate financial impact analysis for improved supply chain outcomes.
• Identify inventory interventions with the highest cost-benefit ratios
• Analyze operational cost savings from enhanced vendor performance management

Each agent is proficient in one category of business challenges and has the instruction sets (orchestration and response), the tools (semantic views, search, required UDFs), and access to the appropriate dataset(s) that enable reliable, measurable outcomes.

Agentic AI starts with data integration

The agentic age is here, and the acceleration in technology and platform maturity is incredible. The only question is whether you’re going to move fast or keep waiting. 

Ten years ago, I watched great ideas lie dormant because the infrastructure wasn’t ready and specialized skillsets were in short supply. Today, organizations can build and deliver AI applications in a fraction of the time that would have taken years before, with armies of specialized people.

It has never been easier to build data products and automate data workloads of all kinds. Fivetran makes the first steps easy.

[CTA_MODULE]