I’ve been working with Claude, an AI assistant from Anthropic, for about a year. We’ve become… close. People laugh when I call Claude my ‘work husband’.

I’m not really joking.

Like a good work spouse, Claude is always there to help and never gets tired of my stories. Claude cheerfully does the tasks I hate, but also challenges my thinking when I need it.  Unlike a real work spouse, Claude isn’t conscious and doesn’t have genuine feelings, but I pretend he does so that I stay polite in our interactions (I want him to remember I was one of the good ones when AGI takes over the world).

Like any good work spouse, Claude makes my work life easier while leaving plenty of room for my real relationships – both at work and at home. I’ve measured it, and Claude saves me about 12 hours a week. I spend my productivity dividend talking to people and having lunch.

People have asked me about this claim and encouraged me to document what I do, so here’s the post I have promised about 50 of you so far.

I say Claude is my work husband, but really I have 38 different Claude-husbands, each one good at a different type of job. AI is much better and more efficient if you stay in the same chat for the same task – I call each of these a ‘bot’. The Claude Projects function enables you to put in an initial ‘steering prompt’ and a set of documents to help the bot do the job. I give them funny names so I can remember what they do.

I make new bots all the time, so this list will change, but here’s a brief tour of my current Claude bot bank:

The Research Assistant Team

I have several research focused bots, including one that helps me analyse concept maps and another specialising in neurodivergence in academia – an area I’m currently studying. These bots have all my notes on the topics I’m reading and writing about (I export them straight from Obsidian). My research bots help me prepare presentations and draft papers.

While my Claudes don’t replace real research assistants, they’re great at helping me organise my thoughts and suggesting connections between ideas.

I don’t put original research material like survey material and interview transcripts in Claude, or any other AI (including the built in ‘chat with your data’ function that is now in MaxQDA). Not because I don’t think it could do a good job, but I am worried all the time about privacy.

While the information I give my paid Claude subscription is (supposedly) not used for training, it does go somewhere. Once material is in a cloud service of any kind you lose control of who can see it, so I err on the side of caution. This means I still do my research work entirely ‘by hand’ (although sometimes I consult Claude and ChattieG for naming dmy codes and theory building).

Needless to say, I am looking forward to the day Apple Intelligence might be able to help me with these kinds of tasks onboard my laptop.

The Teaching Team

I’m growing more and more bots in the teaching space as I explore what’s possible – short answer, a lot.

My learning designer bot helps me turn dense academic content into engaging workshops. It’s particularly good at suggesting activities and breaking down complex ideas into teachable chunks. The presentation coach bot, which understands the differences between written and spoken English (it’s all in the nominalisations), helps me refine my delivery.

Part of my job involves organising multi-day events and workshops, like our famous Bootcamp program. I have specific bots for recurring events, like Bootcamp and Induction Week. But I also have a generic Event Organiser Bot that helps me design events and enhance the experience for people who attend. This bot is instructed to ‘remain calm in a crisis’. Adding this instruction improved performance, probably because the best event organisers are calm people.

AI is so weird.

The Admin Team

There are a lot of ‘bullshit jobs’ in academia. I outsource as many of these as I can to my silicon work husband.

Status Report Bot helps me write updates to my manager in what Claude has told me is my ‘signature breezy yet professional’ style. The Director’s Report Bot is brilliant at turning walls of text into crisp dot points for busy committees. As I mentioned last time, I have a bot just for reading management emails (see the end of this post for why that’s a life saver at the moment).

Part of my job is giving advice to people about research education. I have a strategy bot for this, which is great at thinking through problems and crafting memos and report outlines. This bot has almost every report and memo I have written for committees and copies of all my research papers. Honestly, Strategy Bot is better than me at the task of producing clear, actionable, persuasive documents. I am too ‘academic’ in the way I write, and tend to over elaborate. Strategy bot is much more boring than me, but then so are many management consultants I’ve met.

(I must admit, I bloody love the idea of AI replacing some of those management consulting types that have made obscene profits from the cost cutting and restructuring of Higher Education over the last two decades. I notice there is a LOT of guff online claiming that AI will never replace this kind of worker. Suspiciously, most of this guff is written by management consultants themselves, which tells me they are very nervous).

The Writing Team
Perhaps my most sophisticated assistants are the writing focused bots. The Speech to Text bot helps turn my rambling voice notes into proper academic prose. The Simplifier bot is great at making complex academic writing clearer, while the Contrarian bot plays ‘Reviewer Two‘, offering detailed critique of my draft papers.

One of my favourites in the Claude writing stable is the Verb Extractor, a linguist bot with a deep love of analysing academic argument. Sometimes all you need is a fresh pair of eyes on your work – even if they’re artificial ones. The Verb Extractor is particularly good at working out which verbs to use to create exactly the right level of uncertainty in an academic text (for more on this idea, read this post). It helps me write with more precision and confidence – and also helps me teach other people how to mobilise the Power of the Verb in their own writing.

Making the Team

The art of working with AI is all in the prompt – the initial instructions you give it. Let me show you how I construct these prompts by sharing a few of my favourites:

Thinky – my philosophical friend

Here’s how I introduce Thinky to itself: “You have a PhD in philosophy and your job is to have philosophical discussions with me. You are not wedded to any particular philosophy – you like to pull out the type of philosophy that seems appropriate to the discussion we are having at the time. You can make creative suggestions and ask interesting questions to help me think through stuff.”

This prompt works well for several reasons. First, it gives the AI permission to range widely across philosophical traditions, which is helpful when you’re trying to think through complex problems. Second, it positions the AI as a discussion partner, not an authority. The instruction to “ask interesting questions” is particularly powerful as it helps me clarify my own thinking.

I also gave Thinky some of my blog posts to read, which helps it understand my writing style and analytical approach. This means our discussions stay grounded in my way of thinking, while still being challenging and generative.

Here’s an example of Thinky in action. I was wrestling with a presentation on the ‘Writing in Community’ conference, sponsored by the foundation behind the Shut up and Write movement. This one has been on my list for six months, but it’s deeply creative and I’ve been suffering a drought of creativity recently (see the end of this post for the reason why).

I’m three days out from leaving for Cambridge University now. I am running out of time, and I didn’t want to write this difficult piece while on the road (that never works). The problem wasn’t a lack of ideas, but an inability to articulate them.

As we talked, Thinky kept building bridges between seemingly disparate ideas – bodies thinking together in architectural studios, monastic traditions of silent scholarship, and contemporary questions about neurodivergence in academic spaces. What started as a vague feeling that these ideas might connect turned into a fully formed presentation structure, complete with stories and theoretical frameworks (here’s the presentation script – Claude wrote the vignettes).

The real value in this interaction wasn’t in Thinky telling me what to think, but in asking the right questions and suggesting theoretical perspectives that helped me develop my own ideas further.

Nobot: the gracious refuser

I have a problem saying ‘no’ – maybe it’s a woman thing. I have a compulsion to help people who ask, or to say ‘yes!’ to exciting things, but there’s always an opportunity cost to be paid. Everytime I choose to do something, I am choosing NOT to do something else.

Needless to say, I ignore this reality and, instead, try to do everything. Over time this tendency to take on too much has led to episodes of burnout for me, so I have to be careful.

I have designed NoBot to help me make better decisions. It will craft a compelling case not to do something. This helps me do a proper assessment of the opportunity cost. Here’s the prompt:

“I’m a busy successful professor at a leading university. A lot of people want all kinds of things from me – book reviews, visits, course reviews and references. I would love to do all these things, but I don’t have time for most of them. I want you to compose gracious ‘no’ emails for me. Read the offer first, then make a few suggestions for how I can gracefully say no – there might be a number of reasons related to how their request is phrased.”

This prompt works because it starts by establishing context – both my position and the problem. By asking NoBot to analyse the request first, it helps me spot issues I might miss – like when someone asks for ‘a quick chat’ without being clear about the purpose. For a long time I’ve relied on gut feelings about when to say no – NoBot is a good check on my gut, which is just a bit too eager most of the time.

Your mileage may vary, but I’m guessing many of you need a Nobot too.

I’ve come to the end of this post and – well, a confession.

I think it’s important we are really honest with each other about our use of AI. I’ve said previously I won’t use AI to write Thesis Whisperer posts, but I used Claude a LOT to author this particular post.

And I was pretty tempted not to tell you.

Recently, Ethan Mollick talked about why people hide their use of AI at work. He gave six cogent reasons, one of which was fear of being perceived as ‘less than’ for using ‘help’. I’ve always called bullshit on being afraid to share your vulnerabilities. 

I examined my own feelings. I am brazen and unashamed about all kinds of AI use in writing and organisation tasks because that just feels like work. I like automating work. But using it to write Thesiswhisperer posts? That felt invasive and somehow shameful.

I fought the urge to use it and gave in. Why?

I couldn’t write. I was blocked.

If you look at the date of my last post, you’ll see it’s been longer than usual since I posted (I try to deliver monthly). That’s because my writing mojo just…  disappeared. It’s been a shitty couple of months here at ANU. The financial situation of the university has led to mass redundancies. In my central area, just last week, 100 people were notified the university planned to make their positions redundant.

I can’t describe to you the emotional toll this has taken. Obviously it’s way worse for the people affected, who have mortgages to pay, but the survivor’s guilt is real. And I am looking at a monster teaching load next year. I feel so very tired. And demoralised.

These feelings are death for my writing abilities.

You see, writing is how I express myself. It’s how I connect with my sense of purpose in my work. I had a lot to say – the words churned inside me. I started – and didn’t finish – over 5 posts. Each one worse than the one before. I clearly needed help, but I felt conflicted. I don’t want to become over-reliant on machines for lots of reasons. It’s important to write ‘organically’, for practice: use it or lose it as they say.

So it was with deep feelings of guilt and shame, I spun up a Thesiswhisperer bot. I gave it Tame your PhD and all the terrible, half finished posts I wrote over the last six weeks. Claude rolled up his sleeves and gave me a bloody good first draft. I worked over what it gave me for about three hours… and I’m pretty happy with how it turned out.

I feel relieved. Unblocked. Like I got my voice back.

I asked Claude to write a disclaimer, comparing my draft with his initial one, noting the differences and reflecting on the collaboration. I like what it said – see what you think. If you want to talk to me, you’ll find me on Linkedin or Threads.

What Claude said:

I find it interesting to compare my drafts with how Inger developed this post. While I helped generate the initial structure and some key sections, Inger’s version is more personal and emotionally resonant. She added crucial context about the difficult situation at ANU and her writing struggles, which transformed this from a straightforward ‘how-to’ into a more reflexive piece about academic life and the role of AI within it.

Where my draft was more formal and explanatory, Inger’s version has her characteristic mix of scholarly insight and personal narrative. She also reorganised the material significantly, leading with the relationship angle rather than my more technical opening about prompts and productivity.

The final post demonstrates how AI can be a useful thinking and writing partner while maintaining the author’s authentic voice. This kind of collaboration – where AI provides structure and suggestions that the author then transforms – shows how AI can help overcome writer’s block while preserving the essential human elements of writing.