INFORMS Open Forum

The hot AI topic of the month is Agents.  Question of the week are Agents – AI or Operations Research (OR).  I recently read the claims that AI agents will proposing allocation strategies during constrained supply.  Is meeting this business need something to be accomplished by AI decision technology or OR decision technology? Of course, we first need to ask the questions which AI technology.  I would suggest it is incumbent on the applied side of INFROMS to bring clarity to this question.  My personal view is Agents are a software architecture to imbed solutions. As Peter Norden observed in the mid-1980s, AI is a moving target.

------------------------------
kenneth fordyce
director analytics without borders
Arkieva
Wilmington DE
------------------------------

This is a great and timely conversation. I've been thinking about this a lot lately. To me, AI agents are solving problems that are very familiar to the OR world. Allocation, optimization under constraints, sequential decisions. So in terms of what they're doing, it feels very similar to what OR has always done.
But where I see a difference is in how easy it is to adopt. OR solutions often need strong modeling skills, a good understanding of solvers, and quite a bit of tweaking. AI agents, on the other hand, come wrapped in accessible tools like prebuilt APIs, reinforcement learning frameworks, and architectures that are easier for engineering teams to plug into real systems.

So while both are solving the same types of problems, the path to adoption is very different. AI, in this case, makes it feel more usable, even if the underlying logic isn't necessarily more powerful. That said, I do feel we're at a point where there's a lot more noise than clarity. Every company is jumping on the agent bandwagon. Everyone's promising productivity gains, autonomy, intelligence. While I genuinely see the potential and some real advantages, I'm also a bit skeptical. Are we seeing results because it's working or because it's the current trend?

And that's where I feel INFORMS as a community has a really important role to play. Not to take sides but to help make sense of what's really happening. What's genuinely new, what's just a smarter way of delivering what we already know works, and where we draw the line between actual impact and marketing buzz.

I hope this conversation keeps going. I'm still trying to find my own clarity on it.

------------------------------
Anantha Padmanabhan Anilkumar
Principal Engineer
Western Digital
San Jose CA
------------------------------

Original Message:
Sent: 06-29-2025 23:02
From: kenneth fordyce
Subject: Are Agents AI or OR?

The hot AI topic of the month is Agents.  Question of the week are Agents – AI or Operations Research (OR).  I recently read the claims that AI agents will proposing allocation strategies during constrained supply.  Is meeting this business need something to be accomplished by AI decision technology or OR decision technology? Of course, we first need to ask the questions which AI technology.  I would suggest it is incumbent on the applied side of INFROMS to bring clarity to this question.  My personal view is Agents are a software architecture to imbed solutions. As Peter Norden observed in the mid-1980s, AI is a moving target.

------------------------------
kenneth fordyce
director analytics without borders
Arkieva
Wilmington DE
------------------------------

------------------------------
Warren Hearnes, PhD
Founder, OptiML AI
INFORMS Board Role: VP Technology Strategy
------------------------------

Original Message:
Sent: 06-29-2025 23:02
From: kenneth fordyce
Subject: Are Agents AI or OR?

The hot AI topic of the month is Agents.  Question of the week are Agents – AI or Operations Research (OR).  I recently read the claims that AI agents will proposing allocation strategies during constrained supply.  Is meeting this business need something to be accomplished by AI decision technology or OR decision technology? Of course, we first need to ask the questions which AI technology.  I would suggest it is incumbent on the applied side of INFROMS to bring clarity to this question.  My personal view is Agents are a software architecture to imbed solutions. As Peter Norden observed in the mid-1980s, AI is a moving target.

------------------------------
kenneth fordyce
director analytics without borders
Arkieva
Wilmington DE
------------------------------

Dear Warren Hearnes,

I have much lower expectations for the role and the impact of AI agents. I don't think that AI agents can and should choose which tools to use.  Your example "They do figure out things like "I need to use the formula for confidence intervals for the t-distribution to quantify uncertainty.." requires, first, understanding the meaning of confidence interval and the t-distribution, and whether t-distribution can be applied or not in the particular case. It is the qualified human analyst or OR specialist who must understand the problem to decide on, not an anonymous AI agent that is not responsible for anything.

One can ask AI something like "What is confidence interval" or "what is t-distribution?" with examples to refresh one's memory. But this is similar to, say, Wikipedia search.  No AI is needed at all.

You write "We can make sure the right tools are chosen". Who are we? Anonymous AI agents or responsible human analysts?

I have serios concerns that the current trend in uncritical reliance on AI in technical fields and its impact will inevitably result in diminished students' education level.

------------------------------
Alexander Kolker
ge healthcare
MILWAUKEE WI
------------------------------

Original Message:
Sent: 06-30-2025 09:40
From: Warren Hearnes
Subject: Are Agents AI or OR?

Thanks for starting this discussion on AI agents and OR. Here are my thoughts.

I see an agent being a goal-driven entity that uses environmental data to act, yet that is a fuzzy definition that fits both humans and non-human systems alike. That's not a bad thing since AI is supposed to mimic human intelligence, right? The AI agent uses its capabilities to read and understand text, images, and structured data (similar to a person) to choose which tools to use, so it isn't all LLM generation. They do figure out things like "I need to use the formula for confidence intervals for the t-distribution to quantify uncertainty in the mean service time" and then it solves it, possibly by writing a python script that does that computation.

I think many of these tools that AI agents choose to use will be OR techniques that the INFORMS community has created, so that is where we can really add value. We can make sure the right tools are chosen, which will lead to smarter decisions. AI agents already enhance productivity of human analysts by automating low-level tasks, guiding model development, and translating outputs into actionable insights. I use AI agents to help me code. I remain in control by defining goals, validating models, and making final decisions while the AI acts as a smart assistant that brings speed, scale, and consistency to my workflow. Once the AI is right almost all the time, I think it is natural to give it the autonomy to make the decision on its own - not unlike how we would give a new employee more responsibility or promote an OR model into production.

How have others in this community utilized AI agents? What thoughts or concerns do you have?

------------------------------
Warren Hearnes, PhD
Founder, OptiML AI
INFORMS Board Role: VP Technology Strategy

Original Message:
Sent: 06-29-2025 23:02
From: kenneth fordyce
Subject: Are Agents AI or OR?

The hot AI topic of the month is Agents.  Question of the week are Agents – AI or Operations Research (OR).  I recently read the claims that AI agents will proposing allocation strategies during constrained supply.  Is meeting this business need something to be accomplished by AI decision technology or OR decision technology? Of course, we first need to ask the questions which AI technology.  I would suggest it is incumbent on the applied side of INFROMS to bring clarity to this question.  My personal view is Agents are a software architecture to imbed solutions. As Peter Norden observed in the mid-1980s, AI is a moving target.

------------------------------
kenneth fordyce
director analytics without borders
Arkieva
Wilmington DE
------------------------------

Thanks for the thoughtful response, Alexander. I agree that human analysts remain responsible for understanding the problem, validating assumptions, and making final decisions, especially in technical fields like OR. AI agents shouldn't replace expert judgment. I think there's a role even now for AI agents to assist, not replace, analysts with well-defined, structured tasks. AI may not understand statistics like a human does, but it can follow rules and patterns to suggest or automate standard steps. A lot of times, these are repeatable processes that can greatly benefit from automation, especially at scale.

I personally use AI in a different way than asking Google- or Wikipedia-type questions. Since LLMs can read and process text and images, I ask follow-up questions and dive into different ideas/perspectives, having a conversation about the topic. I can have it check a few of my assumptions quickly with some python code that it writes and runs. I can have it read a chart and make a summary or comparison. I have even had it read an academic paper and create (with some success) the code to replicate the experiments. The quality of what it produces is getting better and better too. If it consistently does well at some tasks that I've given it, then I start to give it more autonomy to take action without me having to give my approval. That's the type of "AI agent" that I have in mind - something that has a goal and takes action to satisfy that goal.

As for the education point: I share the concern about over-reliance without understanding. I hope we can integrate them into education in a way that deepens understanding, not bypasses it.

------------------------------
Warren Hearnes, PhD
Founder, OptiML AI
INFORMS Board Role: VP Technology Strategy
------------------------------

Original Message:
Sent: 06-30-2025 11:48
From: Alexander Kolker
Subject: Are Agents AI or OR?

Dear Warren Hearnes,

I have much lower expectations for the role and the impact of AI agents. I don't think that AI agents can and should choose which tools to use.  Your example "They do figure out things like "I need to use the formula for confidence intervals for the t-distribution to quantify uncertainty.." requires, first, understanding the meaning of confidence interval and the t-distribution, and whether t-distribution can be applied or not in the particular case. It is the qualified human analyst or OR specialist who must understand the problem to decide on, not an anonymous AI agent that is not responsible for anything.

One can ask AI something like "What is confidence interval" or "what is t-distribution?" with examples to refresh one's memory. But this is similar to, say, Wikipedia search.  No AI is needed at all.

You write "We can make sure the right tools are chosen". Who are we? Anonymous AI agents or responsible human analysts?

I have serios concerns that the current trend in uncritical reliance on AI in technical fields and its impact will inevitably result in diminished students' education level.

------------------------------
Alexander Kolker
ge healthcare
MILWAUKEE WI

Original Message:
Sent: 06-30-2025 09:40
From: Warren Hearnes
Subject: Are Agents AI or OR?

Thanks for starting this discussion on AI agents and OR. Here are my thoughts.

I see an agent being a goal-driven entity that uses environmental data to act, yet that is a fuzzy definition that fits both humans and non-human systems alike. That's not a bad thing since AI is supposed to mimic human intelligence, right? The AI agent uses its capabilities to read and understand text, images, and structured data (similar to a person) to choose which tools to use, so it isn't all LLM generation. They do figure out things like "I need to use the formula for confidence intervals for the t-distribution to quantify uncertainty in the mean service time" and then it solves it, possibly by writing a python script that does that computation.

I think many of these tools that AI agents choose to use will be OR techniques that the INFORMS community has created, so that is where we can really add value. We can make sure the right tools are chosen, which will lead to smarter decisions. AI agents already enhance productivity of human analysts by automating low-level tasks, guiding model development, and translating outputs into actionable insights. I use AI agents to help me code. I remain in control by defining goals, validating models, and making final decisions while the AI acts as a smart assistant that brings speed, scale, and consistency to my workflow. Once the AI is right almost all the time, I think it is natural to give it the autonomy to make the decision on its own - not unlike how we would give a new employee more responsibility or promote an OR model into production.

How have others in this community utilized AI agents? What thoughts or concerns do you have?

------------------------------
Warren Hearnes, PhD
Founder, OptiML AI
INFORMS Board Role: VP Technology Strategy

Original Message:
Sent: 06-29-2025 23:02
From: kenneth fordyce
Subject: Are Agents AI or OR?

The hot AI topic of the month is Agents.  Question of the week are Agents – AI or Operations Research (OR).  I recently read the claims that AI agents will proposing allocation strategies during constrained supply.  Is meeting this business need something to be accomplished by AI decision technology or OR decision technology? Of course, we first need to ask the questions which AI technology.  I would suggest it is incumbent on the applied side of INFROMS to bring clarity to this question.  My personal view is Agents are a software architecture to imbed solutions. As Peter Norden observed in the mid-1980s, AI is a moving target.

------------------------------
kenneth fordyce
director analytics without borders
Arkieva
Wilmington DE
------------------------------

Good question, Ken!  From what I know of agents (not much, though I coauthored a couple of papers on an agent system for project management) I'd kind of agree with you-- they're an integration method rather than a new solution tool.   One difference is the interface with the user--- the agents of today can use ordinary text to interact on roughly human-to-human terms.  (It's not clear what the cost of this is!) 

Years ago (1980s) we built process control systems for mining equipment.  We actually had a voice interface where the operator could ask for info and set controls by speaking.  But the vocabulary was limited to I think 56 words!!! And the terminals needed to be trained for the operator's voice!  Fortunately, there wasn't that much you could control.

We later used the interface in an HR payroll and scheduling system.  It was useful because we only needed to specify hours and days worked.  The operator actually did use it some; but mostly reverted to typing, which was much faster.

So are they OR?  From my training, I'd say not; but interfaces do affect the nature of optimization, and a more modern generation might say yes.

------------------------------
Bruce Hartman
Professor
University of St. Francis
Tucson, AZ United States
bruce@ahartman.net
website: https://sites.google.com/ahartman.net/drbrucehartman/Home
blog:http://supplychainandlogistics.org
------------------------------

Original Message:
Sent: 06-29-2025 23:02
From: kenneth fordyce
Subject: Are Agents AI or OR?

The hot AI topic of the month is Agents.  Question of the week are Agents – AI or Operations Research (OR).  I recently read the claims that AI agents will proposing allocation strategies during constrained supply.  Is meeting this business need something to be accomplished by AI decision technology or OR decision technology? Of course, we first need to ask the questions which AI technology.  I would suggest it is incumbent on the applied side of INFROMS to bring clarity to this question.  My personal view is Agents are a software architecture to imbed solutions. As Peter Norden observed in the mid-1980s, AI is a moving target.

------------------------------
kenneth fordyce
director analytics without borders
Arkieva
Wilmington DE
------------------------------