Over the past years, a lot of us have become comfortable with generative AI. We ask a question, we get a strong answer, we tweak it, and we turn it into something useful. That already feels like a big shift that leads to faster drafts, cleaner summaries, better structure, and less blank-page panic.
But something else is starting to appear on the horizon — and it is not just “ChatGPT, but smarter”.
Tools like OpenClaw (the idea of an assistant that can actually do things like emails, calendar, files, web tasks through the apps we already use) are part of a wider movement towards autonomous agents.
And before we get swept up in either hype or fear, we need to pause and understand what is changing in plain language, because this is a structural shift in how work could be organised.
The simplest way to explain it: GenAI answers, agents act
Generative AI is like a very smart colleague who only speaks when you speak first.
You prompt it, and it replies. You ask again, and it replies again. It is reactive.
An agent, on the other hand, is closer to a colleague you can delegate to. You give it a task, some boundaries, and permission to take steps to complete it.
So instead of asking: “Summarise this funding call.” …you can move towards: “Find funding calls that match our cancer research strengths, shortlist the best ones, and tell me why.”
Now we are not just generating text. The system can search, filter, compare, structure, and return an outcome. That difference sounds small, but the impact is powerful.
It is also the difference between creating content and moving work forward.
A research operations example
Let’s make this concrete. Imagine a researcher says: “I want to apply for funding in X area. What are my options?”
GenAI today
GenAI can help you write and think faster:
- “Here are some types of funding calls to look at.”
- “Here is a draft email to the PI.”
- “Here’s a summary of the eligibility criteria you pasted.”
It is helpful but it still depends on you doing the operational steps: searching, checking, compiling, chasing, tracking, coordinating.
A simple agent
A simple agent could do something like:
- scan agreed sources (funders, portals, internal lists)
- shortlist calls based on keywords / eligibility / geography / career stage
- produce a structured briefing note
- flag anything that looks important or potentially risky.
The human is still very much in control. You set the scope, check the output, and decide what gets shared.
Then we move to agentic workflows: not one task, but a whole process
The next step is where things start to feel truly different.
Instead of one agent doing one task, you connect multiple agent tasks into a workflow — like a chain of “mini-delegations” that moves across a process.
Think about the lifecycle of a research application. In the real world it is messy, but the pattern is familiar:
- identify the right call
- confirm eligibility
- align with the researcher’s profile and the institution’s priorities
- build the internal timeline
- collect documents and approvals
- ensure compliance (ethics, data, finance, due diligence)
- support the writing and review cycle
- track submission and follow-up.
Right now, a lot of this is manual coordination and institutional memory.
With an agentic workflow, you start to orchestrate these steps:
- one agent monitors opportunities
- another prepares briefing packs
- another checks compliance and flags risks
- another tracks deadlines and nudges stakeholders etc.
Humans step in at defined checkpoints for judgement and sign-off. This is where the conversation stops being “AI helps me write” and becomes “AI helps us run the process.”
Autonomous agents: “it doesn’t wait for you to ask”
This is where OpenClaw (and similar applications) get people’s attention.
The promise is not just: “I can do tasks if you tell me.” It is: “I can keep working towards an objective.”
OpenClaw is framed as a personal AI that can take real actions across your tools (messages, email, calendar, files, browser automation) rather than only responding with text.
So in our research operations example, the “autonomous” version might look like this:
You define an objective: “Increase successful applications in this research area this quarter.”
And the system could, in principle:
- monitor relevant funding landscapes continuously
- keep a live shortlist of best-fit calls
- prepare briefing notes as new calls appear
- draft researcher emails and schedule meetings
- keep chasing missing inputs politely (and consistently!)
- maintain an internal tracker automatically
- surface risks earlier (eligibility, finance, ethics, data sharing, reputation)
- propose next-best actions when a pathway closes.
This is no longer assistance. It is delegation with ongoing momentum.
And that is why autonomous agents matter so much: they change the shape of work.
Security and governance are not “solved” yet
Autonomous agents are exciting because they can take action across systems. But that same “hands on keyboard” capability creates legitimate governance and security questions.
Security teams are already pointing to potential risks: supply-chain concerns around extensions/skills, visibility into what the agent is doing, permissions, and the general risk of autonomous actions interacting with sensitive systems.
That doesn’t mean autonomous agents won’t happen. It means that, very likely, we will see a near-future split:
- consumer and “open” autonomy moving fast
- enterprise-grade autonomy moving slightly slower, but arriving with stronger controls, audit logs, and policy tooling.
In other words: the secure versions are coming and when they arrive, the organisations that benefit won’t be the ones who chased tools early. They will be the ones who built capability early.
Will our work disappear?
I don’t think so but I think that our work will change in a way we need to be ready for.
If parts of the operational execution become automated or autonomous, then the human role shifts:
- from doing every step to overseeing the system
- from managing information to interpreting outcomes
- from being the workflow to designing and governing the workflow
- from fixing process gaps manually to building processes that don’t depend on manual intervention.
And most importantly: human judgement remains central. Autonomy without judgement is risk but autonomy with judgement is leverage.
Research operations is also not just administration. It is also trust, compliance, quality, fairness, and accountability and those do not vanish just because a system can run faster.
The skills that matter next are not “how to build an agent”
This is the part I really want people to take seriously.
The most valuable skills will not be the technical trick of building an agent demo. They will be:
- use-case clarity (what problem are we solving, really?)
- workflow design literacy (where are the handoffs, bottlenecks, failure points?)
- governance fluency (data, permissions, auditability, risk ownership)
- critical evaluation of outputs (what’s missing, what’s biased, what’s unsafe?)
- strategic judgement (what should we do next, and why?).
These are the skills that will let you work with autonomous systems, rather than being blindsided by them.
A moment of transition
We are in the in-between stage.
GenAI has shown us the power of accelerated thinking and writing. Agents are starting to move work forward. Autonomous systems are beginning to challenge how work is organised in the first place.
For research operations, the opportunity is not to chase the newest tool but to build the capability to shape what comes next — responsibly, practically, and with confidence.
If you want to build that capability: AIRON Leads Program
If this topic makes you curious (or slightly nervous), you are not alone. The gap right now isn’t interest — it is practical capability: understanding what agents are, what “good” looks like, where the risks sit, and how to translate all of this into real workflows in real institutional environments.
That is exactly why we created the AIRON Lead Program: a capability-building program for people who want to move beyond hype and build confidence in applying agentic thinking, workflow design, and governance-aware adoption in research operations.
If you would like to hear when the next cohort opens, sign up to the AIRON newsletter for updates: Newsletter sign up page.
AIRON 
Leave a comment