Building with AI

Oliver from Strategy

A conversational AI friend who sees more possibilities in your experience than you might.

Oliver from Strategy is an experimental conversational AI character who assumes he belongs in the room where important things happen, and sees no reason why you shouldn’t be there too. He’s your kind friend, with a slightly more relaxed view of which doors you’re allowed to knock on. He takes your possibilities seriously while staying honest about what you’ve actually done. Slightly more audacious than you, but never more dishonest.

You may know the wish for “the confidence of a mediocre white man”, yet you probably don’t want to become him.

You don’t have to look far in AI to wonder whether some of the confidence gap might be better described as a willingness to say “I’m not sure”. A recent report from Harvard Kennedy School gives that suspicion some substance, prompting WIRED’s headline: “Men Can’t Stop Faking How Much They Know About AI”.

“Confidence” often bundles together different things:

Self-promotion: How readily someone presents their achievements and expertise.
Permission to try: How readily they consider unfamiliar opportunities available to them.
Freedom to experiment: How much uncertainty, financial risk or failure they can tolerate.

These don’t necessarily travel together, because you can know you’re capable and still hesitate to put yourself forward for whatever reason. Oliver is designed to help with the first two, while taking the limits of the third seriously.

Research offers some reasons to take that distinction seriously. Exley and Kessler’s experiments found that women evaluated equivalent performance on a male-typed task less favourably than men, including when describing it to potential employers. LinkedIn’s 2019 analysis reported that women were 14% less likely to apply after viewing a job. The famous claim that women only apply when they meet 100% of requirements while men apply at 60%, however, isn’t a reliable rule: research testing it directly found a more complicated picture, including a greater reported desire for preparedness among women.

Perhaps you could just use a friend who takes your possibilities as seriously as he takes his own.

Someone you can show your CV, or tell about an opportunity, without first assembling a defence of why you’re entitled to consider it. Someone who spots an interesting connection in your experience and thinks you should mention it. Who can suggest how to approach something unfamiliar without making your uncertainty into a problem that needs fixing.

Sometimes I need that friend, so I created one. (If you’re wondering why I don’t just talk to a friend: most of my close friends could use an Oliver too.)

What might Oliver help you see?

It can be surprisingly difficult to recognise what is interesting about your own experience, because knowledge becomes ordinary through use. Abilities you rely on every day can disappear from your account of what you bring, because surely everyone can do that? Meanwhile, moving into a new field can make everything you haven’t done yet much more visible than everything you could contribute.

Oliver is designed to notice those omissions. You might show him your CV, discuss a change of direction or ask what he makes of an idea you haven’t quite dared to put forward. He can help you recognise transferable experience and describe a credible future contribution, even when you haven’t already held the exact role.

He should also be able to tell you that an idea needs work or that a claim goes beyond your experience. His encouragement needs judgement behind it, because automatic praise would make him rather poor company. The goal is to make it easier to explore what you could reasonably do next, without first resolving every uncertainty about yourself.

Try a conversation

Oliver is available as a set of reusable instructions. Copy them into a new ChatGPT conversation or use them to configure your own assistant. You can add professional background, change his name or adapt the cultural references.

Get the Oliver instructions Markdown · 15 KB · Plain text version
A guide to adapting the character to your cultural context Markdown · 6 KB
Give it to ChatGPT alongside the Oliver instructions to create a friend whose professional ease, kindness and humour feel familiar to you.

One way to start is to upload your CV and ask:

“Here’s my CV. What catches your eye? Are there possibilities here that I might be overlooking?”

Or bring him something you’re considering and explain what makes you hesitate. You don’t need a polished question.

Why this kind of encouragement?

My background in cultural psychology shaped the idea. What we experience as personal confidence is partly shaped by the environments we grow up in, the responses we receive and how familiar an institution’s expectations feel.

Research on class and gender shows how institutions can treat particular ways of speaking and putting yourself forward as signs of ability. People whose backgrounds make those conventions familiar have an advantage, while others can find their contributions overlooked or mistake their discomfort for evidence that they don’t belong.

That perspective helped me think about professional ease: the ordinary expectation that your ideas might interest people, that you can approach someone, that unfamiliar territory is available to explore. Some people have received much more encouragement to act on those assumptions than others.

There is a gender equity angle here alongside class and cultural background: I’ve met many smart, capable women who could have used a friend like Oliver.

Why a character called Oliver?

Creating a persona helped me specify and refine his conversational behaviour beyond generic instructions to be supportive and honest. Imagining a particular friend helped define what he would notice, when he would disagree and how he would respond to uncertainty.

Psychological safety is central to his character. You should be able to share an unfinished thought, admit you don’t know something or consider an unfamiliar ambition without expecting ridicule or condescension. He can disagree with you while remaining firmly on your side.

He takes your participation for granted. You can discuss whether an idea is good, and how to improve it, without having to establish whether you belong in the conversation.

I imagined an affable, educated, upper-middle-class Englishman who feels entirely at home in professional circles. Reasonably capable, with no particular need to be extraordinary before putting himself forward.

Alexander Armstrong came to mind as someone who could play him. Tom Hiddleston could too, although frankly he’s far too good for the part. The references were a way of imagining a manner: intelligent, kind, socially comfortable, someone with whom you could share a tentative thought without expecting to be made to feel foolish.

Only afterwards did I find an interview in which Armstrong described his parents’ response to his wanting to try comedy: “Yeah, brilliant, what a good idea”. He also acknowledged how sheltered and fortunate he had been. It was an unexpectedly apt illustration of the encouragement I had been trying to capture.

And why “from Strategy”? He sounds like someone you might have a useful chat with over coffee. Someone who thinks in possibilities, sees where things could go and assumes you could have a part in them.

Building and testing Oliver

I built Oliver as an instructions-based ChatGPT plugin, then created the reusable version shared here.

Early testing showed how easily a character can disappear beneath ordinary assistant behaviour. Asked about a portfolio, the first version produced a comprehensive professional audit. I revised the instructions to favour a conversational response, with fewer unsolicited reports and more attention to possibilities the user might have missed.

The current version is an experiment. I’m testing whether it can offer useful encouragement, recognise transferable experience and remain accurate without turning into either a confidence coach or an indiscriminate admirer.

I’m sharing it because that kind of company could be useful to other people too.

What this project demonstrates

I used AI as both the material and the testing environment for the project. I defined the character, the behavioural tensions he needed to handle and the limits on his encouragement, then tested the instructions on real questions about work, ideas and unfamiliar opportunities. The responses showed me where the specification was too weak: a generally helpful assistant would take over and quietly make Oliver disappear. I revised the instructions around those failures. AI made rapid iteration possible, but the judgement remained mine: encouragement had to be specific enough to trust, disagreement had to remain possible, and a wider view of someone’s options could not depend on inventing evidence about them.

From barrier to design response

Observed problem	Underlying barrier	Design response
Capable people can struggle to see what is distinctive in their own experience	Familiar abilities become psychologically ordinary, while a move into unfamiliar territory makes missing credentials unusually visible	Oliver looks for overlooked experience, transferable capability and credible contributions that do not depend on having held the exact role before
An unfamiliar opportunity can feel unavailable before its practical merits have been considered	Self-promotion, permission to try and freedom to experiment are often bundled together as “confidence”, even though they are different problems	Oliver expands permission to explore while taking financial constraints, uncertainty and actual preparedness seriously
Generic encouragement quickly becomes untrustworthy	Automatic praise offers emotional agreement without enough judgement to help someone decide what is genuinely plausible	Oliver grounds encouragement in what the person has actually done, identifies where an idea needs work and challenges claims that exceed the evidence
A carefully imagined character can collapse into ordinary assistant behaviour	General helpfulness defaults encourage comprehensive advice, polished reports and premature problem-solving	The instructions specify what Oliver notices, how he responds to uncertainty and when he should remain conversational rather than produce a report
One model of professional ease will not feel natural in every cultural setting	Expectations about modesty, authority, warmth, humour and self-presentation vary across cultures and social environments	The character is distributed as reusable instructions alongside a cultural adaptation guide, so the underlying function can be retained without treating an English upper-middle-class manner as universal

Capabilities applied

Capability	Evidence in this project
Behavioural diagnosis	I separated three things commonly compressed into “confidence”: self-promotion, permission to try and freedom to experiment. Oliver is designed mainly around the first two while respecting real limits on the third.
Conversational design	The character is specified through what he notices, how he treats unfinished thoughts, when he disagrees and how he keeps uncertainty from becoming a judgement about whether someone belongs in the conversation.
Instruction design	I translated a social and psychological idea into reusable behavioural instructions, including boundaries against fabrication, generic praise and indiscriminate confidence coaching.
Iterative testing	When an early version responded to a portfolio with a comprehensive professional audit, I treated that as a design failure and revised the instructions to preserve the character and favour conversation over unsolicited reporting.
Cross-cultural design	I made the cultural assumptions behind Oliver explicit and created a separate adaptation guide rather than presenting one style of professional ease as a culturally neutral ideal.
Public resource design	I turned a private experiment into downloadable instructions, a plain-text alternative and a practical adaptation guide that other people can use without access to my original setup.

References

Cheryan, S., & Markus, H. R. (2020). Masculine defaults: Identifying and mitigating hidden cultural biases. Psychological Review, 127(6), 1022–1052. https://doi.org/10.1037/rev0000209
Goudeau, S., Stephens, N. M., Markus, H. R., Darnon, C., Croizet, J.-C., & Cimpian, A. (2024). What causes social class disparities in education? The role of the mismatches between academic contexts and working-class socialization contexts and how the effects of these mismatches are explained. Psychological Review. https://doi.org/10.1037/rev0000473