B. The Experiential Dimension: What It Feels Like to Work with AI

Purpose: The workshop mostly addresses what Claude can do. This document addresses what it feels like to work with it — the emotional and psychological texture of the experience. These states are real, they affect behaviour, and researchers rarely talk about them directly. Naming them is the first step to navigating them well.

Why this matters for the workshop

Every person in the room has an emotional relationship with AI tools, whether or not they have consciously formed one. That relationship shapes how they use the tools, whether they stick with them, whether they use them well or poorly.

The workshop can address this directly — not as therapy, but as honest craft discussion. A researcher who understands their own experience of AI is better positioned to use it critically than one who assumes their reactions are just personality quirks.

The control and ownership axis

The central experiential dimension of working with AI is how much control and ownership you feel over your work.

This is not a fixed property of the tool — it varies by task, workflow, and how you structure your interaction. But it is the dimension that most powerfully shapes whether working with AI feels good or bad.

High control / high ownership — you feel like the author of the work:

Low control / low ownership — you feel like a passenger:

The uncomfortable truth: both states can produce the same output. The difference is entirely in your relationship to it. Low ownership is not visible in the document — but it is visible in the thesis defence, the peer review, the seminar, the moment someone asks a follow-up question.

The practical implication: structure your use of AI to stay on the high-ownership end. This usually means more specific prompts, more review passes, more iterative dialogue — and less "generate me a full section" in one shot.

The positive states

Joy — the moment when something hard becomes easy.

Batch-processing 40 PDFs in the time it would have taken to open the first one. A grant section that comes out well on the third pass instead of the tenth. A bibliography formatted correctly on the first try. These moments are genuinely pleasurable — and they are not trivial. They represent real time and cognitive load returned to you.

The joy is real. It does not need to be qualified. A researcher who is delighted that Claude drafted their introduction does not need to be reminded that they still have to revise it — they know that. The delight is legitimate.

Curiosity — what else can this do?

Working with Claude reliably generates curiosity in most researchers who engage with it seriously. Curiosity about the tool's limits, about what would happen if you asked differently, about whether it could handle the harder version of the task. This is productive curiosity — it is how people develop genuine fluency.

The curiosity can also become a trap (see below). But in itself it is a good sign: it means the researcher is engaging with the tool as a tool, not just as a widget.

Flow — working with rather than against.

When the workflow is well set up — CLAUDE.md in place, task well-scoped, output format defined — there is a quality of flow to the interaction. Prompts come naturally, outputs need modest revision, the work accumulates. This is the state experienced researchers describe when they say Claude "fits" into how they work. It does not come immediately, and it is fragile — a poorly framed prompt or an unexpected failure mode breaks it. But when it is present, it is real.

The difficult states

Overwhelm — too many tools, too many options, too many updates.

The AI landscape changes faster than any researcher can track. New models, new tools, new best practices, new integration possibilities — and constant noise about all of them. The researcher who felt on top of things six months ago may feel behind today, not because they regressed but because the landscape moved.

This is a structural feature of the current moment, not a personal failing. It affects everyone — including the people writing the breathless "10 ways AI will transform research" threads. Overwhelm is a reasonable response to genuine information overload.

Practical mitigation: Narrow the aperture. You do not need to track the full landscape — you need one or two tools that work for your actual tasks. Claude Desktop plus Claude Code covers most researcher needs. Once those are working, the noise from everything else can be deprioritised.

Anxiety — am I doing this right? am I falling behind?

Two related but distinct anxieties:

The competence anxiety: Am I prompting correctly? Am I missing techniques that would make this better? Is there a way to use this tool I don't know about? This anxiety is partly useful (it motivates learning) and partly counterproductive (it prevents just using the tool).

The position anxiety: Am I falling behind colleagues who use AI more aggressively? Will researchers who use AI more produce more papers, win more grants, advance faster? Should I be doing more?

Both anxieties are widespread and rarely spoken aloud. They drive a pattern of frantic tool-exploration that produces little actual research output — researchers spend time learning AI workflows instead of doing the work those workflows are meant to support.

FOMO — fear of missing out — is the social form of anxiety.

It presents as: "I saw a thread about [new feature / new tool / new workflow] and I need to try it." Or: "A colleague is using [X] and says it changes everything." Or: "I should set up [integration] before I start the next project."

FOMO-driven AI use has a characteristic shape: lots of setup, lots of exploration, limited output. The researcher is always preparing to use AI rather than using it.

The productivity question

What is real productivity, and what is anxiety-driven productivity?

This is the central question the workshop can raise but each researcher has to answer for themselves.

Anxiety-driven productivity has these markers:

Real productivity has different markers:

The distinction matters because AI tools can serve either mode — and they serve anxiety-driven productivity very well. The tool does not know what you actually need. It just does what you ask.

Work-life balance: the specific AI dimension

AI tools introduce specific work-life balance dynamics that did not exist with previous research tools.

The compression benefit is real. Tasks that previously consumed an evening can take an hour. A batch extraction job that would have required a full day can run in a morning. This compression is genuinely valuable and can return time to the rest of life.

The expansion risk is equally real. Because tasks can be compressed, more tasks fill the space. The researcher who finishes a literature review in half the time may find they have simply committed to writing twice as many papers. The tool accelerates the work; the work expands to fill the acceleration.

Accessibility makes boundaries harder. Claude is available at any hour, from any device, for any task. The research day no longer has a natural endpoint. A thought at 10pm can become a two-hour session. This is a choice, but the tool makes it a very low-friction choice.

The "one more prompt" pattern. AI interaction generates a light version of the engagement loop familiar from social media. The next prompt is always just there. A session that was meant to last 20 minutes extends because there is always one more thing to check, one more pass to run, one more question to ask. This is different from the natural stopping points in traditional research (the library closes, the file is saved).

Practical navigation:

A note on authenticity

Underlying many of these states is a question researchers sometimes struggle to articulate: Is this still my work?

This question is worth taking seriously rather than dismissing. The answer is not simply "yes, of course" or "no, it's the AI's." It depends on how the tool was used and what the researcher understands and owns in the result.

The researchers who report the highest satisfaction with AI-assisted work tend to describe it not as delegation but as dialogue — the AI is a thinking partner, a fast executor, a tireless reviewer, but the research questions, the interpretive judgements, and the intellectual ownership remain theirs. They can point to every decision they made, explain every claim they are making, and stand behind the result.

That sense of authorship — of having genuinely made something — is not a luxury. It is connected to the satisfaction that makes research feel worth doing. Tools that undermine it will eventually be abandoned or resented. Tools that support it will be kept.

The question to ask at the end of a session: Did I make something today, or did I watch Claude make something? The first is good. The second is a sign to recalibrate.

For the workshop

These topics can be raised directly — and usually produce recognition and relief in the room. Researchers rarely have permission to admit that AI use sometimes feels overwhelming, anxiety-driven, or alienating from their own work. The workshop can provide that permission.

A useful framing for discussion: What does working with Claude feel like for you right now? Not "what do you use it for" — the experiential question is more revealing and more useful for calibrating what people actually need from the session.

The workshop does not need to resolve these tensions — it can simply name them, show that they are shared, and offer frameworks (the control/ownership axis, the real vs. anxiety-driven productivity distinction) that help researchers navigate them on their own terms.

Related