Jeff Pflueger

Agentic AI in the Enterprise - By Task Type and Business Function

November 3, 2026

I’ve been exploring Agentic AI security.

I’d like to better understand:

1. Who is adopting agentic systems (firmographics - business types, sizes, etc)
2. What business functions the agentic systems are facilitating
3. The specific tasks they are performing

There’s a fourth factor - pain - that I need to understand. I’ll get to that in a later post. The companies using agentic systems - what are the risks and challenges and who has them most acutely. People talk about a TAM - total addressable market for a product - a colleague recently mentioned the subset of the TAM that most interests sales people - the Pain Addressable Market.

What I learned from my research: These are difficult questions to pin down. Partly because everything is moving so quickly with agentic AI and partly because of a lack of great analysis.

My first swipe at this was survey data of business users of agentic systems and AI. What I learned: Executives like to talk about business function (like customer success, etc) using agentic AI, while technical stakeholders like to talk about the task (coding, debugging, etc). So the data are all over the map.

What’s even worse is that surveys and respondents are inconsistent on even the definition of Agentic AI. For instance - some surveys and responders assume a chatbot is AGENTIC because it is like a customer success AGENT. No. Agentic systems autonomously interact with other systems, like a CRM, etc. A chatbot agent just parses language.

So my next swipe at the research is to try to find sources of telemetry data: Anthropic for instance has information about what tasks its agents are asked to perform and also probably the business function.

Well, here too is a challenge - Anthropic for instance is heavily used for coding, which differs in use from, say ChatGPT. But the telemetry data is actual task - what agentic AI systems are doing.

I want to separate the tasks agentic AI is doing from the business function. Task is interesting because from a security context, different tasks involve different systems and have a different risk profile.

Business Function is also extremely important. Different business functions have different types of risk, and also the VALUE of a security solution to manage the inherent security challenges of agentic AI.

So my approach is to ask two different research questions.

1. A top 10 list for Business Function adoption of agentic AI
2. A top 10 list of tasks being performed by agentic AI

Slop alert: Huge caveat here of using Claude to do this level of research - the underlying sources are thin. There is much Claude is likely taking a guess at. That said - here’s what the output is:

</human>

<ai_output>

Table 1: Top 10 by business function

Ranked by Menlo Ventures’ 2025 departmental AI spend, the only behavioral dataset organized by business function. Anthropic and OpenRouter data are added where a task maps clearly to a function.

Table 2: Top 10 by agentic AI task

Ranked by share of Anthropic’s business API traffic, the largest public dataset that classifies real agent and automation traffic by task. API use there is mostly automation rather than human-in-the-loop help: 77% of business API conversations show automation patterns. Overall, coding-related tasks make up 44–52% of business traffic, and the top 10 tasks together make up 32%. Shares are in the business API sample unless noted.

</ai_output>

<human>

What I learned:

• People are confused about what an “agent” is versus “agentic AI”. Many people call a customer success chatbot an Agent. What we are interested in Agentic AI - systems that take autonomous actions and are connected with other systems
• Task use cases: Coding dominates. And there is a long tail of other task types.
• Business Functions - #1 adopter of Agentic AI is Engineering / software development followed by IT Operations.
• There are so many different business uses and task types, that focusing on a narrow category may not be as effective as just simply focusing on the governance and visibility into ALL of them across the enterprise
• The DATA of how agentic AI is being used in the Enterprise becomes a valuable product itself

Sources:

• Anthropic Economic Index – AI’s role in the US and global economy
• Anthropic Economic Index – Uneven adoption (Sept 2025)
• Anthropic Economic Index – Economic primitives (Jan 2026)
• OpenRouter / a16z – State of AI: 100T token study
• Xenospectrum – OpenRouter agent vs. human tokens
• Microsoft – 2026 Work Trend Index: Agents
• Menlo Ventures – 2025 State of Generative AI in the Enterprise
• No Jitter – Menlo spend breakdown and methodology
• SaaStr – 55% of departmental AI spend is coding