About Agentic AI @ KFUPM
A community, not a lab. We exist so that the people at KFUPM working on and with AI agents can find each other — and so that the good work already happening here stops being invisible.
About us
Why this community exists
Agentic AI is moving quickly, and at KFUPM it is being picked up in a dozen places at once — a research group here, a student project there, a pilot inside a department somewhere else. Most of that work is happening in parallel, and most of it is invisible to the rest of campus.
Agentic AI @ KFUPM is a university-wide community that brings together faculty, researchers, students, and practitioners interested in exploring, building, and applying AI agents across disciplines. The idea is simple: bring people together, discover what others are working on, share ideas and resources, and find opportunities to collaborate.
We are not a degree programme, a research centre, or a service desk. We are the layer in between — the regular meeting, the shared reading list, the place you go when you need someone who has already hit the problem you are hitting now.
How we work
- Open by default. Anyone at KFUPM can attend, present, and propose activities.
- Cross-disciplinary on purpose. Sessions are pitched for a mixed audience.
- Practical and honest. We are as interested in what failed and what it cost as in polished results.
- Credit where it is due. Ideas and work stay attributed to the people who did them.
- Low overhead. Light structure, few committees, and a bias toward just running the thing.
Every discipline here has a problem that agents might help with, and a reason they might not. Both are worth hearing.
Foundations
What is agentic AI?
Agentic AI represents the evolution of AI from systems that simply respond to prompts into systems that can perceive, reason, plan, use tools, make decisions, and act autonomously to achieve goals.
A chatbot vs. an agent
A model on its own is a very good text producer: you ask, it responds, the exchange ends. An agent wraps that model in a loop and gives it hands.
| Assistant / chatbot | Agentic system |
|---|---|
| Responds to one prompt | Pursues a goal over many steps |
| You decide each next step | It plans and re-plans its own steps |
| Produces text or code | Takes actions with real effects |
| Forgets between sessions | Keeps state and intermediate results |
| You judge the output | It checks, retries, and reports back |
Anatomy of an agent
Almost every agentic system, however it is branded, is assembled from the same six parts.
A model that reasons
A large language or multimodal model does the interpreting and planning — but on its own it only produces text.
Tools and actions
Search, code execution, a simulator, a database, a lab instrument, an API. Tools are what turn an answer into an effect.
Memory and context
Notes, retrieved documents, prior steps and results, so the agent can work over hours rather than one message.
A control loop
Something that decides what to do next, checks whether it worked, and retries or changes approach. This is what makes it agentic.
Guardrails and oversight
Permissions, budgets, approval steps, logging, and a human who can stop it. Autonomy without these is a liability, not a feature.
Evaluation
A way to measure whether the agent actually succeeds at the task, repeatedly. Without it, demos look far better than reality.
Autonomy is a dial, not a switch
Most useful systems today sit in the middle of this range. Knowing where your system sits is the difference between a helpful tool and an unaccountable one.
| Level | What the system does | Human role | Typical example |
|---|---|---|---|
| Assistive | Suggests; the person acts | Does the work | Code completion, literature summary |
| Delegated task | Completes a bounded task end to end | Defines the task, reviews the result | “Reproduce this figure from the paper” |
| Supervised loop | Plans and executes many steps, pausing at checkpoints | Approves key actions | Multi-step simulation or data pipeline |
| Autonomous, bounded | Runs continuously inside hard limits | Sets limits, audits logs | Monitoring agent that opens tickets |
| Multi-agent | Several agents divide work and negotiate | Designs the system, owns the outcome | Research or engineering pipelines |
The community
Agentic AI @ KFUPM
What the community actually consists of: five recurring activities, a light organizing team, and an open membership list.
Who is in it
Faculty and researchers from across the colleges, graduate and undergraduate students, research and technical staff, and practitioners from partner organizations. Membership means you are on the list and welcome at everything — nothing more, nothing less.
How it is organized
The initiative was launched and is hosted by the Department of Industrial and Systems Engineering in College of Computing and Mathematics, but it is designed as a university-wide community. It brings together faculty members, researchers, students, and practitioners interested in exploring, building, and applying AI agents across departments and research centers. A small volunteer organizing team maintains the calendar, curates the radar, and coordinates meetups and symposia. All other activities, including interest groups, talks, projects, and challenges, are proposed and led by community members.
Community lead: Dr. Alaa Khamis
Faculty sponsor: College of Computing and Mathematics
Hosted by: Department of Industrial and Systems Engineering
The rhythm of the year
- Every monthCommunity meetup
One talk, three to five lightning updates, and open floor. The heartbeat of the community.
- ContinuousResearch & collaboration groups
Small groups meeting on their own schedule and reporting back at meetups.
- Monthly digestResource & research radar
What the community read, tried and recommends — published as a living page plus an email digest.
- Every quarterSymposium
A half-day of keynote, contributed talks, posters and demos, open to the whole university.
- Twice a year (target)Challenge or hackathon
A real problem, mixed teams, a deadline, and something working at the end.
Ground rules
Be generous with credit, careful with other people's unpublished work and data, and respectful of the fact that people come with very different levels of background. Share what you can, ask before redistributing what is not yours, and assume good faith.
Explore. Share. Connect. Build.
Ready to see it in person?
Join the list, then come to the next monthly meetup. That is all there is to it.